Do early intervention services for psychosis maintain their effects after transition to usual/modular care? A systematic review and meta‐analysis

Abstract

Early intervention services (EIS) for psychosis have demonstrated superiority to usual care/modular care (UC/MC) until the end of their delivery. However, the maintenance effects of EIS care after transition to UC/MC are less clear. We aimed to compare these effects vs. UC/MC at least one year after the end of EIS care. This PRISMA and MOOSE‐compliant systematic review searched PubMed, EMBASE, PsycINFO and Web of Science databases and Cochrane Central Register of Reviews, without time or language restrictions. We included studies initially designed as randomized controlled trials (RCTs) comparing EIS vs. UC/MC in patients with early‐phase psychosis, in which both the intervention and control groups were followed for at least 12 months after cessation of EIS care in the intervention group. Co‐primary outcomes were psychiatric hospitalization, duration of hospitalization, and drop‐out at the end of follow‐up (preferably 5 years post‐EIS initiation). Secondary outcomes were severity of total, positive and negative symptoms; quality of life, work involvement, remission, legal offences, antipsychotic use, and suicide attempts. We meta‐analyzed six RCTs with data from 13 papers, including 1,416 individuals (mean age: 23.9 years, females: 36.7%). After 2‐3 years of receiving UC/MC, subsequent to 2‐3 years of EIS care or UC/MC, individuals who originally received EIS care spent less days hospitalized than those continuing UC/MC (n=5, standardized mean difference, SMD=0.128, 95% CI: 0.019‐0.237, p=0.021). However, although we confirmed the superiority of EIS care to UC/MC at the end of the intervention period (except for work involvement and legal offences), the two groups did not differ significantly at 2‐3 years post‐EIS care regarding hospitalization, all‐cause drop‐out, quality of life; severity of total, positive and negative symptoms; work involvement, remission, legal offences, antipsychotic use, and suicide attempts. In summary, EIS care did not maintain its superiority over UC/MC 2‐3 years after its cessation across meta‐analyzable outcomes, except for duration of hospitalizations. These results support the need to further develop and potentially extend full or individualized EIS delivery.

Schizophrenia is one of the most severe and functionally limiting medical disorders, leading to substantial individual, societal and health care burdens ¹ , ² . Outcomes in this disorder, including remission and recovery rates, often remain suboptimal ³ . All‐cause mortality is increased in people with schizophrenia compared to the general population, with the greatest risk in first‐episode (relative risk, RR=7.4) and incident (i.e., earlier‐phase) schizophrenia (RR=3.5) ⁴ . Furthermore, the mean duration of untreated psychosis (DUP) has been estimated as 42.4 weeks worldwide ⁵ , and longer DUP after illness onset has been related to poorer illness outcomes ⁶ , ⁷ , ⁸ .

Early intervention strategies/services (EIS) aim to timely detect psychosis symptoms, reduce DUP, and improve early access to effective treatment during what has been called the “critical period” for intervention effectiveness, which is believed to be the first 2‐5 years following psychosis onset ⁹ , ¹⁰ , ¹¹ , ¹² .

In a meta‐analysis of randomized trials lasting an average of 16.2±7.4 months and including 2,176 patients with first‐episode or early‐phase schizophrenia (mean illness duration: 159.8±125.4 weeks, mean DUP: 79.9±71.1 weeks), EIS care has been associated, at the end of its delivery, with significantly better outcomes than usual care/modular care (UC/MC) ¹⁰ . The improved outcomes included lower treatment discontinuation, less psychiatric hospitalizations, less relapses; more remission, recovery and involvement with school or work; lower severity of total, positive, negative and depressive symptoms as well as general psychopathology; and greater global functioning and quality of life, each evaluated after 6‐24 months of active EIS treatment ¹⁰ .

A limitation of the above meta‐analysis was that insufficient studies and outcome data were available to assess the longer‐term effects of EIS care vs. UC/MC after the former care ended and patients were followed in usual care environments. However, a robust understanding of whether and which benefits of EIS care are sustained beyond the initial active treatment window is essential to inform clinical decision‐making, resource allocation, and long‐term service planning. Such information is further relevant to support or challenge the hypothesis of a “critical period” of 2‐5 years during which the short‐term effects of EIS care would translate into lasting long‐term effects on the above‐mentioned outcomes.

Clarifying the durability of EIS benefits has far‐reaching implications, not only for individual patient trajectories, but also for the sustainability and cost‐effectiveness of mental health systems. This information is, therefore, highly relevant for program developers and policy makers, as current strategies may need to be improved or optimized. This could include the potential need for paradigms of selective extension of EIS care for subgroups of patients or for selected outcomes that may be less durable than others.

Based on the above, we aimed to compare meta‐analytically – across key clinical, functional, and service use outcomes – the impact of EIS care vs. UC/MC both acutely, at the end of the intervention period, and long‐term, i.e. at least one year after its cessation. We hypothesized that EIS care would be superior to UC/MC after transition to the latter for some but not all outcomes.

METHODS

This study was conducted in accordance with the PRISMA 2020 item checklist ¹³ and MOOSE checklist ¹⁴ (see supplementary information). The study protocol was registered on PROSPERO (CRD42023488103).

Literature search

PubMed, EMBASE, PsycINFO and Web of Science databases and Cochrane Central Register of Reviews were searched by two independent researchers. The following search terms were applied from inception until February 18, 2025: (schizophrenia OR schizoaffective OR schizophreniform OR psychosis OR psychotic) AND (recent‐onset OR “recent onset” OR “first episode” OR first‐episode OR early‐phase OR “early psychosis”) AND (intervention OR treatment OR management OR integrat* OR multimodal OR assertive OR specialized OR “COAST” OR “JCEP” OR “LEO” OR “OPUS” OR “OTP” OR “PIANO” OR “STEP" OR “RAISE” OR “RAISE‐ETP” OR “Valencia”) AND (“follow up” OR “follow‐up” OR “longitudinal” OR “long term” OR “long‐term” OR “maintenance” OR “prospective”).

Papers identified were screened as abstracts and, after the exclusion of those not meeting our inclusion criteria, full texts were assessed for eligibility. Searches were completed with manual backward and forward reference searching (looking at previously published papers and those citing the included studies). Any discrepancies were resolved through discussion with the first author.

Any missing data that helped assess eligibility were obtained by contacting the corresponding and/or first and last authors of the studies. Furthermore, at least one author from the included trials was contacted to gather additional information on evaluated outcomes in their studies. Authors from all included studies except one responded to our request and are included as co‐authors in this paper.

Inclusion/exclusion criteria

We included: a) randomized controlled trials (RCTs) comparing EIS/coordinated specialized care using an integrated/multimodal program, irrespective of previous service provision, with non‐specialized UC/MC; b) for adolescents aged ≥12 years and/or adults; c) diagnosed with first‐episode or early‐phase (study‐defined) psychosis or schizophrenia‐spectrum disorders (schizophrenia, schizoaffective disorder, schizophreniform disorder, delusional disorder, psychotic disorder not otherwise specified, or affective psychosis); d) in which both the intervention and the control groups were followed for at least 12 months after cessation of EIS care in the intervention group.

We excluded: a) trials initially randomizing patients to EIS care versus a step‐down/less intense maintenance treatment, and b) cross‐over trials in which EIS care was not stopped, or its intensity was not reduced. For overlapping studies and samples, the largest sample for each outcome was included (allowing ≤50% overlap), but included studies were independent regarding the outcomes.

Outcomes

We used the following co‐primary outcomes at follow‐up after transition to UC/MC: a) psychiatric hospitalization, measured by proportion of people hospitalized, and duration (days) of hospitalization; b) drop‐out, defined by being lost to follow‐up (previously referred to as all‐cause treatment discontinuation).

Secondary outcomes included: a) quality of life (study‐defined); b) severity of total, positive and negative symptoms (study‐defined); c) work involvement (% of patients with study‐defined work); d) remission (% of patients who met the study‐defined definition); e) relapse (% of patients who met the study‐defined definition); f) legal offences (% of patients with at least one study‐defined offence event); g) antipsychotic use; and h) suicide attempts.

We selected the closest time‐point to five years after treatment was started (e.g., after two years of EIS care or UC/MC and three years of follow‐up, or after three years of EIS care or UC/MC and two years of follow‐up).

The same primary and secondary outcomes were assessed at the end of the initial period of EIS care compared with UC/MC.

Data extraction

Three researchers independently extracted data from the included studies into a database (a Microsoft Excel spreadsheet). The following variables were extracted: first author, publication year, country, program, patient age (mean±SD, range), sex (% females), sample size (baseline, follow‐up; total, EIS, UC/MC), race/ethnicity, trial duration (active EIS and post‐EIS follow‐up, in years), psychiatric diagnoses, outcome results, variables evaluated in sub‐analyses and meta‐regression analyses, and quality assessment.

Data synthesis

For the meta‐analyses of continuous outcomes, we calculated standardized mean differences (SMDs) with their 95% confidence intervals (CIs). For categorical outcomes, we calculated risk ratios (RRs) with their 95% CIs. Intent‐to‐treat (ITT) data were used when‐ ever possible.

Expecting high heterogeneity, random‐effects meta‐analyses were conducted whenever at least three studies provided meta‐analyzable results ¹⁵ . Publication bias was assessed by Egger’s test ¹⁶ , complemented by the “trim and fill” method to correct for the presence of missing studies when risk of publication bias (i.e., small sample bias) was detected. Any correction was based on the assumption that the effect sizes of all the studies were normally distributed around the centre of a funnel plot; in the event of asymmetries, the “trim and fill” method adjusted for the potential effect of unpublished studies.

Heterogeneity among study point estimates was assessed using Q statistics. The proportion of the total variability in the effect size estimates was evaluated by the I² index ¹⁷ and considered statistically significant when p<0.05. I²>50% is typically considered an indication of high variability in effect size estimates.

We conducted sub‐analyses for the three co‐primary outcomes to estimate the association between the outcomes and a) region (Europe vs. the US vs. rest of the world), b) blinding of outcome assessments, c) analyzed data type (ITT vs. observed cases), d) EIS care including vocational/educational rehabilitation, e) EIS care including family therapy, f) EIS care including crisis response teams, g) EIS care including social skills training, h) EIS care including fidelity monitoring, and i) quality assessment results (studies with low risk of bias vs. studies with some concerns).

We conducted meta‐regression analyses for the three co‐primary outcomes whenever at least four studies provided relevant data, estimating the association between the outcomes and a) sample size, b) number of sites, c) trial duration (total from baseline, EIS treatment, and post‐EIS follow‐up), d) number of EIS components, e) number of visits, f) proportion of patients with DSM/ICD‐defined schizophrenia, g) baseline functioning (measured by the Global Assessment of Functioning, GAF ¹⁸ ), h) severity of positive symptoms at baseline (subscores from the Positive and Negative Syndrome Scale, PANSS ¹⁹ ), i) severity of negative symptoms at baseline (subscores from the PANSS), j) presence of at least one prior hospitalization, k) mean patient age, l) % of females, m) % of White people, n) median DUP duration, o) % of patients not followed from on‐EIS treatment into the off‐EIS treatment study (magnitude of attrition or potential non‐random attrition), and p) % of patients with all‐cause discontinuation in the post‐EIS study.

An additional sensitivity analysis was conducted excluding the Early Assessment Service for Young People with Psychosis (EASY) trial ²⁰ , ²¹ , as participants were recruited from a low‐resource setting and exposed to EIS care before the RCT. We also compared the raw values of continuous and categorical outcomes in the EIS and UC/MC groups (in the on‐EIS and off‐EIS phases). For continuous outcomes, we considered the most frequently used instrument to evaluate each outcome and required at least two studies to meta‐analyze the results.

All analyses were two‐sided, with alpha=0.05, using Comprehensive Meta‐analysis (CMA) V3 ²² .

Risk of bias assessment

Cochrane Risk of Bias tool (RoB2) ²³ was used to assess whether each study had a high, low or unclear risk of bias in each of the following six domains: random sequence generation, allocation concealment, blinding of participants and study personnel, blinding of outcome assessments, incomplete outcome data, and selective outcome reporting.

The overall risk of bias was classified as low if none of the domains was rated as high risk and at least three domains were rated as unclear risk; as unclear if no or one domain was rated as high risk, but at least four were rated as unclear risk; and as high if more domains were rated as high or unclear risk.

RESULTS

The literature search yielded 3,802 non‐duplicate citations, which were screened for eligibility. Ultimately, 13 papers from six studies were included: the OPUS trial ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ ; Lambeth Early Onset (LEO) ³¹ ; Optimal Treatment Project (OTP) ³² ; Recovery After an Initial Schizophrenia Episode ‐ Early Treatment Program (RAISE‐ETP)/NAVIGATE ³³ ; Specialized Treatment Early in Psychosis (STEP) ³⁴ ; and EASY ²⁰ , ²¹ (see Figure 1 for PRISMA flow chart and Table 1 for characteristics of the included studies).

Figure 1.

PRISMA flow chart. EIS – early intervention services

Table 1.

Main characteristics of included studies

Study	Program (country)	Age, mean±SD (range)	% females	Sample size, baseline (EIS; UC/MC)	Sample size, follow‐up (post‐EIS; UC/MC)	Duration in years, total (EIS or UC/MC; all UC/MC)	Diagnosis (% schizophrenia)	Risk of bias
Bertelsen et al 30	OPUS (Denmark)	26.6±6.4 (18‐45)	41	547 (275; 272)	301 (151; 150)	5 (2; 3)	67	Low
Chang et al 20	EASY (Hong Kong)	20.8±3.3 (15‐25)	60	154 (82; 72)	143 (76; 67)	5 (3; 2)	84.2	Some concerns
Chang et al 21	EASY (Hong Kong)	20.8±3.3 (15‐25)	60	154 (82; 72)	143 (76; 67)	5 (3; 2)	84.2	Some concerns
Gafoor et al 31	LEO (UK)	Median: 25	35.4	144 (71; 73)	99 (45; 54)	5.5 (2; 3.5)	72	Some concerns
Hansen et al 29	OPUS (Denmark)	26.6±6.4 (18‐45)	41	547 (275; 272)	164 (84; 80)	20 (2; 18)	67	Some concerns
Hastrup et al 24	OPUS (Denmark)	26.6±6.4 (18‐45)	41	547 (275; 272)	301 (151; 150)	5 (2; 3)	67	Low
Nordentoft et al 27	OPUS (Denmark)	24.0±4.4 (18‐35)	34	63 (34; 29)	63 (34; 29)	5 (2; 3)	89.9	Some concerns
Pollard et al 34	STEP (US)	22.7±5.1 (16‐45)	18.8	117 (60; 57)	99 (54; 45)	7 (2;5)	29	Some concerns
Posselt et al 25	OPUS (Denmark)	27.1±6.3 (18‐45)	41	547 (275; 272)	301 (151; 150)	5 (2; 3)	67.2	Low
Robinson et al 33	RAISE‐ETP (US)	23.0±5.1 (15‐40)	27.5	404 (223; 181)	120 (63; 57)	5 (2; 3)	56	Some concerns
Secher et al 28	OPUS (Denmark)	26.6±6.4 (18‐45)	41	547 (275; 272)	347 (181; 166)	10 (2; 8)	67	Low
Sigrúnarson et al 32	OTP (Norway)	25.4±4.6 (18‐35)	38	50 (30; 20)	45 (28; 17)	12 (2; 10)	76	Some concerns
Stevens et al 26	OPUS (Denmark)	26.6±6.4 (18‐45)	41	547 (275; 272)	547 (275; 272)	5 (2; 3)	67	Low

EIS – early intervention services, UC – usual care, MC – modular care

Altogether, 1,416 individuals were included: 741 in the original EIS intervention group and 675 in the control group. The sample size of each study (including both intervention and control groups) ranged from 50 to 547 individuals. The mean age of the total sample was 23.9 years (range: 20.8‐26.6), and 36.7% (range: 18.8‐60.0%) were female. The duration of the studies ranged from 5 to 20 years, with all of them providing data after 5 years from the start of the RCT, except for one trial ³² that provided data after 10 years (total mean study duration: 5.8 years). All studies had two years of EIS care vs. UC/MC and three years of follow‐up, except for one with three years of EIS care vs. UC/MC and two years of follow‐up ²⁰ , ²¹ (mean EIS intervention: 2.17 years; mean post‐EIS intervention: 4.0 years).

Meta‐analytic results

Outcomes at the end of EIS care

Compared to UC/MC, at the end of the EIS period, EIS care was associated with a lower proportion of patients psychiatrically hospitalized (RR=0.908, 95% CI: 0.828‐0.997), fewer days hospitalized (SMD=0.153, 95% CI: 0.007‐0.299), and decreased drop‐out (RR=0.644, 95% CI: 0.552‐0.751) (see Table 2 and Figure 2). EIS care was also associated with significantly lower severity of positive symptoms (SMD=−0.197, 95% CI: −0.389 to −0.006) and negative symptoms (SMD=−0.310, 95% CI: −0.517 to −0.103) (see Table 2 and Figure 3).

Table 2.

Short‐term outcomes of early intervention services

Outcome	N (EIS)	N (UC/MC)	Point estimate			z score	p	Test for heterogeneity
			SMD/RR	95% CI				Q	I2	p
% hospitalization	599	546	0.908	0.828	0.997	–2.030	0.042	2.091	0.000	0.554
N. days hospitalized	376	365	0.153	0.007	0.299	2.056	0.040	1.917	0.000	0.383
Drop‐out	659	603	0.644	0.552	0.751	–5.600	0.000	0.599	0.000	0.963
Positive symptoms	407	365	–0.197	–0.389	–0.006	2.015	0.044	0.000	0.900	0.983
Negative symptoms	407	365	–0.310	–0.517	–0.103	2.937	0.003	0.505	0.000	0.473
Work involvement	407	365	1.053	0.894	1.240	0.617	0.537	0.004	0.000	0.950
Legal offences	291	282	0.847	0.454	1.579	–0.524	0.600	1.149	0.000	0.563
Antipsychotic use	407	365	1.147	1.006	1.307	2.044	0.041	0.176	0.000	0.675
Suicide attempts	407	365	0.560	0.318	0.984	–2.017	0.044	2.392	58.201	0.122

EIS – early intervention services, UC – usual care, MC – modular care, SMD – standardized mean difference, RR – risk ratio

Figure 2.

Short‐term and long‐term categorical outcomes of early intervention services (EIS). For % hospitalized, drop‐out, legal offences and suicide attempts, risk ratio (RR) <1 indicates benefits of EIS vs. usual care/modular care (UC/MC); for antipsychotic use and work involvement, RR>1 indicates benefits of EIS vs. UC/MC.

Figure 3.

Short‐term and long‐term continuous outcomes of early intervention services (EIS). Negative values of standardized mean difference (SMD) indicate benefits of EIS vs. usual care/modular care (UC/MC).

Use of antipsychotics was more common in individuals on EIS care at the end of the EIS period (RR=1.147, 95% CI: 1.006‐1.307). The number of suicide attempts was lower in the EIS vs. UC/MC group (RR=0.560, 95% CI: 0.318‐0.984). However, EIS and UC/MC groups did not differ in work involvement (p=0.537) or legal offences (p=0.600) (see Table 2 and supplementary information).

Heterogeneity across studies was not statistically significant (all p>0.05).

Outcomes ≥1 year after cessation of EIS care

Individuals who had originally received EIS care spent less days psychiatrically hospitalized after 5 years (including 2‐3 years of UC/MC in both groups) than those receiving UC/MC from the beginning (n=5; EIS: N=691, UC/MC: N=618; SMD=0.128, 95% CI: 0.019‐0.237, p=0.021) (see Table 3).

Table 3.

Long‐term outcomes of early intervention services

Outcome	N (post‐EIS)	N (UC/MC)	Point estimate			z score	p	Test for heterogeneity
			SMD/RR	95% CI				Q	I2	p
% hospitalization	691	618	0.9	0.787	1.029	–1.548	0.122	1.83	0.000	0.767
N. days hospitalized	691	618	0.128	0.019	0.237	2.302	0.021	1.022	0.000	0.906
Drop‐out	741	675	1.009	0.942	1.081	0.269	0.788	5.687	12.083	0.338
Quality of life	226	218	0.306	–0.059	0.670	1.645	0.100	7.435	73.099	0.024
Total symptoms	220	207	–0.168	–0.358	0.023	1.726	0.084	1.293	0.000	0.524
Positive symptoms	291	282	–0.012	–0.176	0.152	0.149	0.886	0.900	0.000	0.638
Negative symptoms	250	266	–0.012	–0.185	0.161	0.136	0.892	0.914	0.000	0.633
Work involvement	287	277	0.969	0.695	1.351	–0.186	0.852	0.101	0.000	0.951
Remission	298	292	1.071	0.901	1.274	0.780	0.436	0.086	0.000	0.958
Legal offences	351	339	0.619	0.222	1.714	–0.919	0.358	5.216	42.486	0.157
Antipsychotic use	291	282	0.991	0.905	1.085	–0.198	0.843	0.816	0.000	0.665
Suicide attempts	291	282	0.739	0.304	1.798	–0.666	0.505	5.022	60.175	0.081

post‐EIS – individuals receiving UC/MC after an initial EIS care, UC/MC – individuals receiving usual or modular care from the beginning of the trial, SMD – standardized mean difference, RR – risk ratio

EIS and UC/MC groups did not differ in the proportion of patients hospitalized (n=5; EIS: N=691, UC/MC: N=618; RR=0.90, 95% CI: 0.787‐1.029, p=0.122) or drop‐out (n=6; EIS: N=741, UC/MC: N=675; RR=1.009, 95% CI: 0.942‐1.081, p=0.788) after EIS care cessation (see Table 3 and Figure 2).

EIS and UC/MC groups did also not differ in severity of total, positive and negative symptoms; quality of life, work involvement, remission, legal offences, antipsychotic use, or suicide attempts (all p>0.05) (see Table 3 and Figures 2, 3).

Heterogeneity across studies was statistically significant for quality of life (Q=7.435, I²=73.099, p=0.024), indicating high variability in the effect size estimates, but not for the other outcomes.

Sensitivity analyses and publication bias

The proportion of patients with psychiatric hospitalization, duration of hospitalization, and drop‐out were not significantly influenced by region of study conduct, blinding of outcome assessment, analyzed data type; EIS care including vocational/educational rehabilitation, or family therapy, or crisis response teams, or social skills training, or fidelity monitoring; or by study quality (all p>0.05) (see also supplementary information).

The proportion of patients with psychiatric hospitalization, duration of hospitalization, and drop‐out were not significantly influenced by sample size, number of sites, trial duration, number of treatment components, number of visits, proportion of patients with DSM/ICD‐defined schizophrenia, baseline functioning, baseline severity of positive or negative symptoms, prior hospitalizations, mean patient age, sex, ethnicity/race, median DUP, % of patients not followed into the off‐EIS treatment study, or % of patients with all‐cause discontinuation in the post‐EIS study (all p>0.05) (see also supplementary information).

Potential publication bias was found in three of the evaluated outcomes: positive symptoms (t=71.729, p=0.009), negative symptoms (t=19.945, p=0.032), and work involvement (t=48.441, p=0.013). However, adjusted results via trim‐and‐fill method did not affect the significance of the results (see supplementary information). Excluding the EASY trial did not change the significance of any of the results (see supplementary information).

Of the findings in the short term showing significant superiority of EIS care vs. UC/MC, only the significantly lower number of days hospitalized (47.2 ± 34.5 days vs. 58.4 ± 29.5 days, mean difference: 11.2 ± 3.0 days, p = 0.0002) remained significant after long‐term follow‐up off‐EIS (68.6 ± 20.3 days vs. 75.5 ± 20.1 days, mean difference: 7.0 ± 3.4 days, p = 0.017). All other significant findings on EIS treatment vs. UC/MC lost significance during the follow‐up off‐EIS period.

This loss of significant group difference favoring EIS care included the previously lower proportion of patients hospitalized (on‐EIS: 50.4% vs. 58.1%, p=0.008; off‐EIS: 31.9% vs. 34.5%, p=0.342), lower number of drop‐outs (on‐EIS: 26.6% vs. 35.1%, p=0.001; off‐EIS: 58.3% vs. 59.0%, p=0.813), higher antipsychotic use (on‐EIS: 68.5% vs. 59.5%, p=0.015; off‐EIS: 61.4% vs. 66.9%, p=0.181), and lower proportion of patients with suicide attempts (on‐EIS: 8.3% vs. 28.0%, p=0.002; off‐EIS: 7.6% vs. 6.0%, p=0.6314) (see also supplementary information).

Finally, excluding trials that only contributed to long‐term but not short‐term data did not change the non‐significant differences between the prior EIS and prior control groups in any of the outcomes (see supplementary information).

Quality assessment

Altogether, five studies (38.5%) were at low risk of bias and 8 studies (61.5%) were scored as having some concerns (see Table 1 and supplementary information).

DISCUSSION

To our knowledge, this is the first meta‐analysis to compare the acute effects of EIS care to the maintenance effects of this care after transition from EIS to UC/MC in people with early‐phase psychosis.

We confirmed the superiority of EIS care vs. UC/MC from the prior meta‐analysis of the acute effects of EIS ¹⁰ in all but two outcomes, despite having only six programs with maintenance data, instead of 10 in that prior meta‐analysis ¹⁰ . However, after cessation of EIS care, individuals who had received 2‐3 years of EIS only spent less days hospitalized than individuals receiving UC/MC from the start, while none of the other meta‐analyzed outcomes differed anymore at 2‐3 years after EIS cessation. Specifically, after cessation of EIS care, its effects were not superior anymore to those of UC/MC regarding risk of psychiatric hospitalization, drop‐out; total, positive or negative symptom severity; quality of life, work involvement, symptomatic remission, legal offences, antipsychotic use, or suicide attempts.

Importantly, there was a mixed picture of why the previously significant differences in favour of the patients randomized to EIS care were lost compared to the control group during off‐EIS follow‐up. There was some long‐term improvement in both the prior EIS and control groups with regard to the proportion of patients hospitalized, with a somewhat greater reduction in the prior control group. Regarding being on antipsychotic treatment, the prior EIS group had an about 10% reduction and the prior control group had an about 10% increase, levelling off the difference in the post‐EIS period. As to the proportion of patients with suicide attempts, the numbers remained roughly similarly low in the prior EIS group, but decreased by about 75% in the prior control group. Finally, drop‐outs increased in both groups during follow‐up, but much more in the prior‐EIS group (by about half, compared to the about one‐third increase in the control group).

Thus, while EIS care seems beneficial in the short term for reducing hospitalization and improving symptomatic and functional outcomes ¹⁰ , advantages are lost during ongoing usual care. Although we focused on results after five years to favour consistency, the OPUS trial, using register‐based data of the entire study population after 20 years of follow‐up, similarly did not find differences between EIS care and UC/MC in symptomatic, functioning, use of services, or mortality outcomes ²⁹ .

How can this loss of sustained benefits across almost all meta‐analyzable outcomes be explained? Several methodological, illness‐related and treatment‐related reasons may apply, informed also by the mixed trends in the frequency of good vs. bad outcomes. First, since less studies provided long‐term follow‐up data, the non‐significant findings might simply be due to underpowered analyses. However, we analyzed data from the same studies at the end of EIS care, finding that, even in this smaller dataset of six studies, benefits were replicated in the on‐EIS study phase versus the full data set of 10 studies ¹⁰ .

Second, in observational follow‐up studies, lower functioning patients are more likely retained in clinical services, causing a potential floor effect in the post‐EIS phase. This possibility is hinted at by the greater drop‐out rate in the post‐EIS than in the UC/MC group, with a possibility of better responding and functioning patients being lost from the post‐EIS group during follow‐up. Indeed, in the RAISE‐ETP study, only after statistical methods were used that accounted for data not missing at random did the prior EIS group show maintenance of significant benefits over the initial UC group at approximately 3‐5 years post‐EIS cessation ³³ .

Third, it may be that antipsychotic non‐adherence, which is common in people with early‐phase psychosis ³⁵ , leads to psychotic relapses, and that the effects of relapses uniformly attenuate treatment benefits ³⁶ , ³⁷ , independent of earlier treatment gains due to EIS care ³³ . Fourth, relatedly, it is generally acknowledged that currently available antipsychotics are symptom‐based interventions that do not alter the underlying disease process or mechanisms, so that, over time and with treatment gaps and symptomatic worsening, initial benefits of EIS care are washed out.

Fifth, there may be illness phase‐specific risks, such as suicide risk, that are highest in the early phases of psychosis ³⁸ , ³⁹ , whereby the overall lower risk later on can level off detectable differences between EIS care and UC/MC. In fact, we observed this pattern regarding the proportion of people with suicide attempts, in that the protective effect of active EIS care was apparent in the short term, and remained at that relatively lower level in the post‐intervention EIS group, while the risk in UC/MC decreased, arriving at a similarly lower level during follow‐up.

Sixth, the care generally accessed and provided as part of UC/MC may be suboptimal, lacking in consistent antipsychotic coverage (e.g., use of long‐acting injectables, LAIs ⁴⁰ , ⁴¹ ), or lacking consistent psychosocial interventions, and this suboptimal care over years could also wash out initial gains made by EIS care.

Taken together, based on the above, it seems that the “all or nothing” approach to EIS, related to the hope for a “critical period” during which EIS “seals the deal”, is insufficient. Rather, individualized step‐down or step‐up treatment approaches may be needed. This strategy is supported by gains that were maintained when EIS care was extended at full or even partial intensity ⁴² , ⁴³ , ⁴⁴ . As part of these approaches, subgroups will need to be identified that require more or less interventions, also regarding specific components of the EIS approach.

Some previous studies randomized patients after two years of EIS care to either three more years of EIS or UC/MC, and found some benefits on work alliance and patient satisfaction (but not hospitalization or functional outcomes) ⁴² , and symptom remission ⁴³ . Conversely, in the only RCT comparing EIS care with step‐down treatment (although similarly intense as UC/MC and focused on medical management), patients receiving an additional year of EIS care had better outcomes in negative symptoms, depressive symptoms, general psychopathology, global functioning, independent living skills, work productivity, relationships and compliance, than those managed by step‐down psychiatric care ⁴⁴ , ⁴⁵ , ⁴⁶ . Furthermore, in an uncontrolled trial, gains made after two years were sustained or increased with EIS care at a lower level after 3 years ⁴⁷ .

While some guidelines recommend continued treatment only for the first two years after a first episode of schizophrenia ⁴⁸ , ⁴⁹ , ⁵⁰ , results from our meta‐analysis indicate that not only antipsychotic treatment, but also EIS care should be extended beyond two years to maintain their effects, although the intensity of specific psychosocial interventions could possibly be reduced or individualized.

Although we could not confirm sustained impact after cessation of EIS care, individuals who had received that care spent less days hospitalized than those receiving UC/MC throughout, although effect sizes were small. Our results are consistent with a naturalistic, 10‐year follow‐up using a matched historical control, in which EIS care significantly reduced the duration of psychiatric hospitalizations in the long term (odds ratio, OR=1.29) ⁵¹ . Notably, costs were lower with EIS than UC/MC ⁵² , particularly due to reduced inpatient stays ⁵³ and emergency department visits ⁵⁴ . So, reducing hospitalization duration is important. According to this finding, EIS care may still be beneficial in the long term, and limited statistical power may have prevented us from observing a sustained effect in other outcomes, suggesting that longer‐term post‐EIS studies are required, and that evaluating long‐term results from other RCTs conducted to date and with short‐term results available (e.g., Valencia ⁵⁵ or PIANO ⁵⁶ ) could be beneficial.

None of the numerous tested meta‐regressor factors and sub‐analyses helped deconstruct heterogeneity. It is still possible for some interventions to improve particular outcomes or maintain effects in patient subgroups. However, study‐level meta‐analyses of bundled interventions are insufficient to answer these questions, calling for individual patient data meta‐analyses. Precision psychiatry highlights individual variability, and the need to stratify interventions according to individual characteristics may improve targeted outcomes ⁵⁷ , ⁵⁸ , ⁵⁹ . This may help detect subgroups of individuals requiring an adaptation in the duration of the interventions, specific intervention contents, or higher‐intensity interventions ⁵ . Further research should focus on better understanding the sources of heterogeneity in the acute and long‐term response to EIS care.

This study has some limitations. First, with six RCTs and 1,416 subjects, the number of studies and subjects was relatively small, affecting the statistical power of our analyses. Second, we did not have sufficient data to evaluate other outcomes, including depressive symptoms, living independently, recovery, relapse; concurrent obesity, diabetes mellitus, metabolic syndrome, or cardiovascular disease; clozapine use, or treatment resistance. Third, patient and treatment characteristics were heterogeneous; we tried to explore the relevance of this heterogeneity in meta‐regression analyses, but without finding any relevant effects, yet statistical power was again a limiting factor.

Fourth, UC/MC in the included RCTs may have consisted of care that is better and more comprehensive than real‐world UC/MC, as indicated by the number of treatment elements and intensity. On the other hand, UC/MC may also have been insufficient, causing a floor effect for all outcomes. Fifth, only one study from a low‐mental health resource location was included ²⁰ , in which a recruitment bias toward those with lower functioning was found. Finally, capable and willing to consent individuals may not be representative of the real‐life psychosis population.

In conclusion, benefits of EIS care do not seem to be sustained in the years following initial treatment, except for days hospitalized. These results support the need to further develop and potentially extend, or to differentiate and improve EIS care and its delivery. Since specific components of EIS care may have sustained effects beyond the intervention period, a more comprehensive model of EIS may be necessary to test the full extent of the “critical period” hypothesis.

Future trials evaluating integrated models are warranted to assess whether long‐term outcome improvements can be achieved through sustained or stepped interventions, or coordinated interventions that dynamically respond to patient trajectories and needs.