20-Year Prospective, Sequential Follow-Up Study of Heterogeneity in Associations of Duration of Untreated Psychosis With Symptoms, Functioning, and Quality of Life Following First-Episode Psychosis

Changes in models of health care provision for psychotic illness typically evolve incrementally or emerge heuristically (1). In the latter regard, a recent meta-analysis confirmed the superiority of early intervention services over conventional models of care for first-episode psychosis in terms of several symptomatic and functional indices of outcome over periods extending up to 2 years following a first psychotic episode (2). In addition, our recent study indicates that early intervention services are associated with health economic advantage in terms of incremental net benefit in a real-world setting (3).

While justification for early intervention services is multifarious (2, 4, 5), a major driver is evidence that longer duration of untreated psychosis (DUP) is associated with poorer outcome and the corollary that shorter DUP is associated with better outcome (6–8). However, although meta-analyses have confirmed the statistical robustness of these associations (9, 10), a major clinical issue for early intervention services is the extent to which associations between reduction in DUP and improvement in outcome endure longitudinally across the lifetime course of illness. Additionally, there have been continuing concerns regarding whether DUP may, at least in part, constitute not a direct predictor of outcome but rather be confounded with premorbid features of psychotic illness (9, 11, 12) or lead-time bias (13).

Meta-analysis indicates that the evidence base is most extensive during the early years following first-episode psychosis, although such relationships with DUP may be evident for individual clinical indices of outcome during up to a mean of 8 years after the first psychotic episode (9). Because subsequent studies, involving diverse first-episode psychosis and early intervention services cohorts using different indices of outcome over varying periods of follow-up, have produced inconsistent results (14, 15), clarification of these issues requires prospective, systematic studies of epidemiologically representative first-episode psychosis incidence cohorts with follow-up of at least 10 years (14, 16). Additionally, such studies should utilize a broad and consistent range of psychopathological and functional indices of outcome and adopt a statistical approach that allows the course of associations between DUP and outcome to be examined systematically at several time points along what can be decades of illness trajectory.

We previously investigated, in separate sequential studies, associations between DUP and outcome, independent of confounders, in an epidemiologically representative first-episode psychosis cohort at the first psychotic episode and at 4, 8, and 12 years (16), and recently we described initial findings on outcome in this cohort at 20 years post-first-episode psychosis (17). While these sequential studies utilized a diverse range of indices of outcome, each involves a “core” set of assessments for the domains of psychopathology, functioning, and quality of life. Application of linear mixed-model analyses now allows, for the first time, determination of the extent to which prospective associations of DUP with differing domains of outcome are consistent or vary across the trajectory of psychotic illness over a 20-year period.

Methods

Study Design, Cohort Formation, and Management

We conducted prospective, sequential follow-up studies of an epidemiologically representative first-episode psychosis incidence cohort in Ireland. Between February 1995 and February 1999, we screened all referrals to all inpatient and outpatient services in a Dublin catchment area with a baseline population of approximately 165,000 individuals. We formed an epidemiologically complete first-episode psychosis sample (N=171) designed to reflect a real-world clinical setting (16, 17) by including all individuals in our cohort who were age 12 or older and diagnosed with first-episode psychosis using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) (18). First-episode psychosis was defined as the first presentation with acute psychotic symptoms to any psychiatric service for patients who, if they had been prescribed antipsychotic medication prior to presentation, had been receiving antipsychotic treatment for no more than 30 days. Thereafter, during a period when early intervention services for first-episode psychosis had not yet been introduced, patients received treatment with antipsychotics (as outlined in detail in the Results section) in a milieu of inpatient hospitalization, outpatient psychiatric review, nurse home visits, psychological interventions, occupational therapy, social work and family supports, and day center attendance, as considered appropriate.

Approvals for this prospective study were obtained at intervals from the Saint John of God Hospitaller Ministries Research Ethics Committee. All participants included in the analyses provided written informed consent for their participation at baseline and at each follow-up point after study procedures had been fully explained to them. Consent for participation of family members in order to determine the DUP and premorbid social adjustment at baseline was also sought. We adhered to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines for cohort studies (19).

Tracing and Re-Recruitment Procedures

At each follow-up point, we contacted cohort members either directly by telephone or letter or via their general (family) practitioner or clinical team. Contact and general practitioner data collected at the time point before each follow-up were used to enable tracing and re-recruitment.

Assessment of DUP and Baseline Variables

Following a period of stabilization (up to 72 hours) after service contact, DUP was evaluated using the scale established by Beiser and colleagues (20). We conducted separate interviews with participants and their family members, resolved any inconsistencies found between accounts, and determined the time (in months) from first evidence of psychotic symptoms to first presentation to psychiatric services and initiation of treatment with antipsychotics.

At baseline, we recorded age and years of education and categorized participants according to SCID-I diagnosis (nonaffective psychosis, affective psychosis), sex (male, female), living alone status (yes, no), and lifetime drug or alcohol abuse (yes, no). The Premorbid Social Adjustment scale (21) was completed in consultation with participants’ family members to measure social functioning between ages 5 and 11, as described previously (16), given that later years can coincide with the onset of psychosis or its prodrome. If family interview consent was not given, these data were recorded as missing.

Outcome Measures Used at Baseline and Follow-Ups

At baseline and at follow-ups at 6 months and at 4, 8, 12, and 20 years, we collected data on psychopathology, functioning, and quality of life using the same instruments, as described previously (16, 17): total and subscale scores on the Positive and Negative Syndrome Scale (PANSS) (22) were used to assess total, positive, negative, and general psychopathology; functionality was assessed using the Global Assessment of Functioning Scale (GAF) (18); and quality of life was evaluated using the Quality of Life Scale (QLS) (23).

Statistical Analysis

To establish whether participant dropout was random over time and not likely to bias the results, we grouped participants by the number of follow-ups in which they participated and analyzed group differences using SPSS, version 26. Two-way contingency table analyses were run on baseline binary variables, and one-way analyses of variance and Kruskal-Wallis tests were run on nonbinary baseline variables.

We ran longitudinal trajectory analyses using Stata/SE, version 14.2, to test whether the binary and nonbinary baseline variables could be predictors of symptoms, functioning, and quality of life over the study period. Linear mixed models were used because they utilize all available data, accommodate missing values by avoiding sample bias (from omitting participants) and estimation bias (from imputation), effectively handle covariates that change over time, and model time effects flexibly (24). An unstructured variance-covariance matrix of the random effects was specified, and the restricted maximum likelihood method was used for model fitting. For these analyses, DUP was transformed to a factor variable with four levels corresponding to the quartiles of this baseline variable. Rather than performing logarithmic transformation of DUP values to address normality violations, the quantile approach adopted here can provide more accurate estimates for the skewed distributions typical of DUP, and skewedness of DUP may reflect meaningful heterogeneity (25, 26). In the case of each outcome variable, initial models were run that included DUP without and then with other baseline variables, all with a six-level time variable. On including other baseline variables, nonsignificant variables were dropped, and reduced models for each outcome variable were run to include the significant baseline variables with time, DUP quartile, and a DUP quartile-by-time interaction term.

An analysis of the time effect that compared successive time-point differences along the longitudinal trajectory, involving a series of reverse adjacent contrasts, was run using the Stata “ar.” contrast operator. Then, an analysis of the DUP quartile effect, involving a series of reference category contrasts that compared differences between the first quartile and the second, third, and fourth quartiles, was run using the Stata “r.” contrast operator. The criterion for statistical significance of all tests was set at a p value <0.05.

Results

The mean age of the 171 participants at study entry was 29.1 years (SD=12.0). Ninety-nine (58%) participants were male, and 61 (36%) were in full-time employment at baseline. SCID-I diagnoses at baseline were as follows: schizophrenia (N=87, 51%), schizophreniform disorder (N=14, 8%), delusional disorder (N=13, 8%), bipolar I disorder with psychotic features (N=26, 15%), major depressive disorder with psychotic features (N=10, 6%), psychotic disorder not otherwise specified (N=4, 2%), substance-induced psychosis (N=12, 7%), and psychotic disorder due to a general medical condition (N=5, 3%). Given the small numbers for several individual diagnoses, reflecting the real-world diversity of first-episode psychosis in this cohort, these were pooled conventionally (27, 28) into nonaffective psychoses (N=135, 79%) and affective psychoses (N=36, 21%). This allowed us to integrate diagnoses into our analyses in a manner that is better aligned with continually evolving schizophrenia spectrum-dimensional perspectives (29); on this basis, the nonaffective psychosis category included 118 (87%) participants with schizophrenia and schizophreniform disorder, together with delusional disorder and psychosis not otherwise specified, which are increasingly considered to have a spectrum-dimensional relationship to schizophrenia (30–32), and 17 participants (13%) with substance-induced psychosis and psychotic disorder due to a general medical condition. The affective psychosis category included 26 participants (72%) with bipolar I disorder with psychotic features and 10 (28%) with major depressive disorder with psychotic features.

Of the 171 participants, outcome measures were provided by 170 participants at baseline, 116 at the 6-month follow-up, 136 at the 4-year follow-up, 113 at the 8-year follow-up, 136 at the 12-year follow-up, and 80 at the 20-year follow-up (time points 1–6). At the 8-year follow-up, 11 participants were deceased, one was in a persistent vegetative state, and 46 were lost to follow-up. At the 12-year follow-up, 18 participants were deceased, one was in a persistent vegetative state, three declined assessment, and 13 were untraceable. At the 20-year follow-up, 20 participants were deceased, one was in a persistent vegetative state, 50 declined assessment, and 20 were untraceable. The decrease in sample size from the 12-year follow-up to the 20-year follow-up was largely because of compliance with the European Union General Data Protection Regulation, enacted in Ireland in 2018, and its impact on research ethics committee approval of processes for subject tracing, renewal of consent, and re-recruitment. For results of participant dropout analyses, see Table S1 in the online supplement. Across all baseline variables, we found no significant differences between participant dropout groups.

At first-episode psychosis, medication data were available for 142 of the 170 participants (84%); of these, 95 (67%) were initially prescribed a first-generation antipsychotic and 47 (33%) were prescribed a second-generation antipsychotic; none were receiving either a combination of both or no antipsychotic treatment. At the 4-year and 8-year follow-ups, medication data were not available. At the 12-year follow-up, medication data were available for 133 of the 136 participants assessed (98%); of these, 15 (11%) were receiving a first-generation antipsychotic, 69 (52%) were receiving a second-generation antipsychotic, three (2%) were receiving a combination of both, and 46 (35%) were not receiving antipsychotic treatment. At the 20-year follow-up, medication data were available for each of the 80 participants assessed (100%); of these, six (8%) were receiving a first-generation antipsychotic, 47 (59%) were receiving a second-generation antipsychotic, one (1%) was receiving a combination of both, and 26 (32%) were not receiving antipsychotic treatment.

Results of linear mixed-model analyses are presented statistically without and with inclusion of significant baseline variables in Table 1. The 20-year trajectories of each outcome variable by DUP quartile are shown in Figure 1A–D and Figure 2A and B. Four broad trajectories were apparent.

Trajectory 1 was a steep, curvilinear decline in scores (i.e., reduction in symptoms) that was particularly evident over 6 months to 4 years and typically leveled off by 8 years, as indicated by significant overall effects of time. Longer DUP quartile was associated with higher scores (i.e., increase in symptoms) consistently across the 20-year period, as indicated by a significant overall effect of DUP quartile in the absence of DUP quartile-by-time interactions. This trajectory for the effect of DUP was evident similarly for PANSS total, positive, and general scores, both without and with inclusion of significant baseline variables (Table 1, Figure 1A, B, and D).

Trajectory 2 was a more gradual decline in scores (i.e., reduction in symptoms), as indicated by a significant overall effect of time. Longer DUP quartile was associated with higher scores (i.e., increase in symptoms) consistently across the 20-year period, in a generally parallel manner, as indicated by a significant overall effect of DUP quartile in the absence of a DUP quartile-by-time interaction. This trajectory for the effect of DUP was evident for PANSS negative score, both without and with inclusion of significant baseline variables (Table 1, Figure 1C).

Trajectory 3 was a steep, curvilinear increase in scores (i.e., improvement in outcome) that was particularly evident over 6 months to 4 years and then leveled off by 8 years, as indicated by a significant overall effect of time. Longer DUP quartile was associated with lower scores (i.e., deterioration in outcome) across the 20-year period, as indicated by a significant overall effect of DUP quartile. However, this effect of DUP quartile varied with time, being minimal at baseline, consistent over 6 months to 4 years, and less consistent thereafter, as indicated by a significant DUP quartile-by-time interaction. This trajectory for the effect of DUP was found for GAF score, both without and with inclusion of significant baseline variables (Table 1, Figure 2A).

Trajectory 4 was a less steep, curvilinear increase in scores (i.e., improvement in outcome) that leveled off by 4–8 years, as indicated by a significant overall effect of time. Longer DUP quartile was associated with lower scores (i.e., deterioration in outcome) consistently across the 20-year period, in a generally parallel manner, as indicated by a significant overall effect of DUP quartile. However, this effect of DUP quartile varied with time, being consistent from baseline to 4 years and more variable thereafter, as indicated by a significant DUP quartile-by-time interaction. This trajectory for the effect of DUP was found for QLS score with inclusion of significant baseline variables, although temporal variability was not present without inclusion of these variables (Table 1, Figure 2B).

For results of reverse adjacent contrast analyses of time factor and reference category contrast analyses of DUP quartile factor across PANSS, GAF, and QLS scores, see Table S2 and Table S3 in the online supplement.

Discussion

Main Findings

In this study, we utilized linear mixed-model analysis to examine associations of DUP with symptoms, functioning, and quality of life in first-episode psychosis sequentially across six time points over a period of 20 years. In general, we found that shorter DUP was associated with an enduring, beneficial overall effect on outcome over time. However, heterogeneity was apparent in the time course of these effects and in the extent to which they were affected by baseline variables, including indices of premorbid functioning. The rarity of similar studies underscores the importance of these results in relation to models of first-episode psychosis health care provision. Because it is more likely that someone experiencing first-episode psychosis will receive standard care rather than early intervention services, our results may be more generalizable than those from studies using early intervention services samples.

Meta-analyses across studies at a given time point indicate that longer DUP is associated with poorer outcome across given indices, primarily in proximity to first-episode psychosis (9, 10), with a paucity of studies involving individual time points beyond a decade thereafter (14, 16, 17). To our knowledge, this is the first study to systematically examine DUP-outcome relationships in a first-episode psychosis incidence cohort at multiple, sequential time points over a 20-year period; hence, the results are not directly comparable with previous research. Nonetheless, our data appear to be inconsistent with the recent proposal (13) that the effect of DUP on outcome is confounded materially by lead-time bias, because such bias would predict loss of DUP-outcome relationships in the long-term post-first-episode psychosis. Issues relating to sample and methodology in the study conducted by Jonas and colleagues (13) recently received detailed consideration in the Journal (33, 34). There are differences between our study and that of Jonas and colleagues (13) that may contribute to differences in findings. Principally, our sample included typical community as well as hospital presentations and, again typically, was not confined to individuals with a schizophrenia spectrum diagnosis. Also, patients received a broad range of community interventions, in addition to pharmacotherapy. Furthermore, our finding that DUP-outcome relationships endured in our non-early-intervention services sample across 20 years should be juxtaposed with previous research suggesting that the effects of early intervention services may diminish over time and could be largely accounted for by earlier follow-up effects (35, 36). These issues indicate both the critical importance of disentangling these relationships and the methodological complexities involved in doing so.

Heterogeneity Across Trajectories of Outcome

Although total, positive, and general symptoms declined over the early years of the 20-year period, the effect of DUP was evident throughout, and it endured when baseline variables were taken into account. Early diminution in positive symptoms, primarily over the first 6 months, together with slightly delayed diminution in general symptoms, most likely reflects typical responsivity to antipsychotic drugs (37). Nevertheless, the lower levels of positive and general symptoms sustained over 4–20 years still varied in relation to DUP. In contrast, negative symptoms declined more slowly and steadily across this period. Yet the effect of DUP was again evident throughout and endured when baseline variables were taken into account. This slow and steady diminution in negative symptoms most likely reflects interplay between their generally poor responsivity to antipsychotic drugs and the long-term naturalistic course of illness over a 20-year period (38), throughout which diminution in negative symptoms still varied in relation to DUP.

Although functionality increased over the early years of the 20-year study period, its relationship to DUP was not consistent throughout, being less evident at baseline, emerging rapidly at 6 months, and then becoming less prominent but sustained through the 20-year follow-up; this effect of DUP endured when baseline variables were taken into account. Overall functionality is a complex interaction between occupational and social functioning and symptoms (39). Thus, for shorter DUP, the initial rapid increase in functionality may in part reflect the initial rapid decrease in positive symptoms as a result of responsivity to antipsychotics, in association with less prominent negative symptoms. Conversely, for longer DUP, the more gradual increase in functionality may in part reflect the more gradual decrease in negative symptoms, which particularly relate to poor functioning (38), in association with less prominent positive symptoms.

Although quality of life also increased over the early years of the 20-year period, its relationship to DUP differed from that of functionality, being evident at baseline and sustained throughout the 20 years; taking baseline variables into account revealed some variability over later years. Quality of life in first-episode psychosis is also a complex construct that is influenced by several factors, primarily DUP and negative symptoms (40). The present findings indicate that this pernicious effect of DUP on quality of life at first-episode psychosis endures across a 20-year period in a manner that most closely parallels the effect of DUP on negative rather than positive or general symptoms.

Explanatory Processes

Finding DUP to be differentially associated with outcome trajectories across several domains suggests that mechanistic explanation is unlikely to be unitary and more likely involves both biological and psychosocial aspects. We suggest that consideration of biopsychosocial models may assist in understanding the heterogeneity in associations between DUP and different outcome domains identified over 20 years.

While initial proposals that DUP reflects an active, morbid process have not been sustained by either neuroimaging (41, 42) or neuropsychological (43, 44) research, longer DUP has recently been associated with greater loss of hippocampal volume integrity (45) and of functional connectivity, surface area, and cortical thickness across neuronal networks (46). Although the reproducibility of such changes and any longitudinal relationships to symptomatology and functionality remain to be clarified, they may be relevant to the present findings. While dopaminergic hyperfunction is evident not only during the early course of psychotic illness but also during the prodromal period for patients who subsequently progress to clinical psychosis (47, 48), its long-term trajectory has yet to be investigated.

However, dopaminergic hyperfunction has been equated with psychological processes thought to be impaired in psychosis, such as salience attribution (48). Each patient’s interpretation of his or her psychotic experiences may contribute to the longitudinal impact of DUP on outcomes. Additionally, psychological factors can influence the extent to which psychosis impairs functioning and quality of life (49, 50). Early treatment and effective response to antipsychotics may give participants with shorter DUP the psychological space in the initial phase of their recovery to either make meaning from, accept, and actively cope with their psychotic experiences, or even to forget and move on from them. Variations in such processes may contribute to the differentiating trajectories reported, particularly in relation to functionality and quality of life. Further investigation of such a biopsychosocial model (51) may help to better clarify the present heterogeneity in associations between DUP and different domains of symptomatic and functional outcome.

Strengths and Limitations

We recruited a first-episode psychosis sample epidemiologically representative of the diversity of psychotic illness, and this cohort was established prior to early intervention services in the study’s catchment area, providing data from participants with a wide range of DUP. This prospective study involved an exceptionally long-term follow-up data set, with the same wide range of outcomes measured longitudinally over many sequential time points. It included investigators who participated from initiation through the 20-year follow-up to ensure consistent oversight. Our application of linear mixed-model analysis optimized the yield from the study and did not substantiate concerns that associations between DUP and outcome might be materially confounded with baseline variables, such as premorbid features (9, 11, 12), or reflect lead-time bias (13).

As in most longitudinal studies over such an extended period, there was inevitably some attrition in the cohort, although this was offset in part by the statistical approaches adopted. While the domains of outcome assessed were consistent throughout, the instruments applied were typical of usage more than two decades ago, and more incisive instruments have been developed since. For example, the GAF is limited by its inability to distinguish between different domains of functioning (occupational, social, and psychological) and how discordance in symptom severity and functionality influences its validity. In addition, clinical judgment was involved in assessments using both the QLS and the PANSS, and although across this 20-year period each incoming assessor was trained by his or her predecessor using a consistent “handover” protocol, some variations between assessors cannot be excluded. Also, affective symptoms and cognition were not assessed. While we controlled for many baseline variables in our analyses, there may be other factors that influenced the relationships between DUP and the outcomes assessed. As in most naturalistic studies of this type, it was infeasible to collect data on paths to treatment, service engagement, treatments received, including use of clozapine, or participants’ degree of adherence to treatment over the 20-year period. As a result, we were unable to determine how these factors may have affected the relationships reported. Future studies should seek to clarify these issues. Also, because some outcome measures changed rapidly over the early phase, assessments should be enriched during this period to better define their time course.

In summary, although trajectories of associations between DUP and outcome can vary between domains of symptoms, functionality, and quality of life, our finding that associations endure across a 20-year period has implications for models of health care provision. While our study was not a controlled evaluation of the effectiveness of early intervention services, the findings nonetheless imply that efforts to reduce DUP (52) may have clinical benefits that could last for at least 20 years after first-episode psychosis.