Predictors of relapse and engagement in care one year after ending services in an urban safety net coordinated specialty care program for first episode psychosis

Abstract

Objective:

This study aimed to identify risk factors for relapse (psychiatric emergency department visits or hospitalization) and lack of follow-up with outpatient psychiatric care in the 12 months after ending services in an urban safety net coordinated specialty care (CSC) program for first episode psychosis (FEP).

Methods:

The study population (n=143) were individuals with FEP who had any CSC care between 2014 to 2021. To identify risk factors for relapse and follow up after exit, multivariable logistic regression was performed using data from electronic health records and linked insurance claims data.

Results:

Individuals with any emergency department visit or hospitalization 12 months prior to ending CSC (aOR=4.69, 95% CI 1.78–12.34) and those who were using cannabis at last CSC contact (aOR=4.06, 95% CI 1.56–10.56) had a higher risk of relapse after ending CSC services. Cannabis use at last contact was also associated with lower rates of outpatient psychiatric follow-up (aOR=0.32, 95% CI 0.12–0.94), while CSC duration in months had a small positive association with post-CSC psychiatric follow-up. There were no differences in relapse or follow-up by race or ethnicity, primary diagnosis, or medication usage.

Conclusions:

Prior relapse during CSC predicted relapse in the 12 months after ending CSC services, but not outpatient follow up. Cannabis use predicted both a higher rate of relapse and a lower rate of follow up after ending services. There were no differences by race or ethnicity in our sample, suggesting that once individuals engaged in FEP care there were no evident disparities in the observed outcomes.

1. Introduction

Schizophrenia is a leading worldwide cause of disability and early mortality (Olfson et al., 2015; Vos et al., 2017).The first episode of psychosis (FEP) represents a crucial time for intervention. Coordinated specialty care (CSC) is a model of care that provides intensive, time limited, multi-disciplinary services for patients with FEP. CSC consists of individual medication management, resilience focused individual psychotherapy, family interventions, supported employment and education, and peer support (Mueser et al., 2015). Individuals who received care in CSC programs demonstrated improved symptoms and quality of life compared to treatment as usual in the Recovery After Initial Schizophrenia Episode (RAISE) trial in the United States (Kane et al., 2016). There was some evidence of ongoing benefit three years later, although with limited long term follow up (Robinson et al., 2022). International trials have similarly shown that early intervention is superior to treatment as usual (Craig et al., 2004; Petersen et al., 2005). However, longer term treatment effects have waned particularly among more clinically complex and socially marginalized populations (Gafoor et al., 2010; Hansen et al., 2023; Secher et al., 2015), and extending the duration of early intervention has not resulted in long-term benefit (Albert et al., 2017; Chang et al., 2017). Apart from some consensus guidelines and qualitative research, there is little consensus on what kind of care should be delivered and to what specific sub-populations after CSC (Hyatt et al., 2022). Customizing interventions to higher risk subgroups could deliver more tailored treatment for patients leaving FEP programs.

Clinical factors and substance use have been shown to influence clinical course after FEP. In the United Kingdom, relapse after leaving early intervention services was predicted by inpatient hospitalization during the program (Kam et al., 2015) as well as ongoing substance use at discharge (Puntis et al., 2018). It is unclear how applicable these findings are to the United States given vast differences in aftercare with many patients in the UK managed by primary care doctors as opposed to specialty psychiatric care after leaving FEP programs (Ahmed et al., 2015), which is not the norm in the US. In the US, while substance use is a well-known risk factor for poorer outcomes in first episode as well as chronic psychosis (Marino et al., 2020; Moore et al., 2007), RAISE did not demonstrate changes in substance use rates over the duration of the CSC (Cather et al., 2018), and no studies have investigated the effect of substance use on post-CSC outcomes. Cannabis use is of particular concern given well established links between use, higher symptom burden, and higher rates of hospitalization in first episode psychosis (Patel et al., 2016; Schoeler et al., 2016b) as well as in more chronic psychotic illnesses (Moore et al., 2007).

While minoritized patients with psychosis have higher rates of hospitalization in treatment as usual settings (Rost et al., 2011), there were no differences in outcomes between non-White and White patients assigned to CSC in RAISE-ETP. Some have suggested that CSC may have promise in alleviating racial disparities in psychosis by addressing modifiable risk factors for poor outcomes in a clinical trial setting (Oluwoye et al., 2018). However, Black individuals assigned to CSC used less family therapy than White individuals, which could result in poorer long-term outcomes given family therapy’s proven efficacy in reducing relapse (Bighelli et al., 2021). Data from New York’s CSC program OnTrackNY showed that non-Hispanic Black and Hispanic individuals had more inpatient days compared to White patients in the six months before and after discharge, suggesting racial disparities may persist in real world settings (Humensky et al., 2021). In the United Kingdom, individuals from minoritized racial and ethnic groups and from more impoverished areas were also more likely to relapse after discharge from early intervention services (Ahmed et al., 2015; Griffiths et al., 2023; Kam et al., 2015). Given the well-known intersection of income, wealth, and race and ethnicity (Bailey 2017), it is also concerning to note that a secondary analysis of the RAISE data found that CSC only improved symptoms and quality of life for the top quartile of socioeconomic status with negligible impact for the bottom 75% (Bennett and Rosenheck, 2021).

The present study aims to address the dearth of US-based studies investigating outcomes after CSC through a retrospective analysis of an early psychosis program at an urban safety net community health system. Risk factors for poor outcomes after leaving the program were investigated in order to understand who should be prioritized for more intensive services to maintain benefits after the end of CSC. Additionally, the diverse setting allowed for investigation into whether individuals from minoritized racial and ethnic groups were at particularly high risk.

2. Methods

2.1: Description of the program

The study took place at an urban academic community health system serving over 140,000 patients annually, consisting of two acute care hospitals and fifteen community clinics providing primary and specialty medical care as well as inpatient and outpatient mental health and substance use treatment. The patient population is disproportionately non-white and non-English speaking compared to state averages, and approximately 65% are publicly insured or uninsured. The health system has been an early adopter of innovative models of psychiatric care including CSC (Hyatt et al., 2021).

The FEP program began in 2014 and runs in accordance with the coordinated specialty care (CSC) model for first episode psychosis (NIMH, 2013), with the following components: individual, family and group therapists, psychiatry, case workers, certified peer specialists, and supported employment and education specialists. These final two positions were added in 2018 after further support was obtained from the Commonwealth of Massachusetts. Individuals could be referred from inpatient units, outpatient psychiatry, primary care, or the community, and had to be within three years of their first episode of psychosis. CSC services were generally limited to three years, but duration was flexible based on individual needs. Aftercare was available within the system including primary care, general outpatient psychiatry, and a specialized outpatient behavioral health home for individuals with chronic psychotic illness (Tepper et al., 2017).

2.2: Data sources and study sample

This analysis included individuals who received care for first episode psychosis in the health system’s CSC, defined as having had an intake and at least one treatment visit in the program from August 1, 2014 through December 31, 2022. Entrance into the CSC was defined as their first evaluation with a CSC clinician. CSC exit was defined by a transfer to another service or clinic, the patient discontinuing services, or the patient being declared lost to follow up as indicated by the clinic staff writing a termination note indicating the patient was no longer receiving services. This was extracted by manual review of the health record, and duration of time enrolled in CSC was defined as the number of days between entrance and exit.

Data were collected from electronic health records (EHR) and were extracted at the person-month level in the 12 months before and after exit from the program. Partial claims data (at least one month) was available for 55% of the sample and was merged with EHR data. Manual chart review was used to extract variables not available in automated data extraction.

A total of 216 patients were referred to the first episode psychosis program between 2014 and 2021. Nineteen of those referred were never seen for an initial evaluation, and seven patients were still enrolled in CSC at the time of analysis and so were excluded. An additional 32 were seen for an intake but did not receive any treatment visits because they were deemed to not have first episode psychosis, they declined services, or were immediately lost to follow-up. Three patients died by suicide and one by an accident while enrolled, all within eighteen months of entry, and were excluded. This is a mortality rate similar to other FEP cohorts (Bertelsen et al., 2007). Finally, 11 patients were immediately lost to follow-up on discharge (ex. moved out of the country) and so were excluded. Thus, a total of 143 of the 216 referred patients had follow-up data available and were included in the analysis. There were no significant differences in race (p=0.26), language (p=0.36), age (p=0.32), insurance status (p=0.38), substance use at intake including cannabis (p=0.92) and tobacco (p=0.42), and baseline functional indicators such as living arrangements (p=0.092), education (p=0.83), or current employment or school (p=0.061) between those who received services and the group of all referred patients. There was a difference by sex (p=0.02) with women being less likely to engage in services after referral than men.

The study was approved by the Cambridge Health Alliance Institutional Review Board.

2.3: Outcome variables

The primary outcome of interest was a composite metric of relapse in the 12 months after leaving the CSC program, defined as any emergency department (ED) visit for a psychiatric complaint or any inpatient psychiatric admission in a given month, a commonly used and valid measure of psychosis relapse in the literature (Humensky et al., 2021; Olivares et al., 2013). The secondary outcome was any follow-up with a psychiatrist or with a therapist 12 months after ending CSC services given their inclusion in best practice recommendations about aftercare when leaving CSC (Daley et al., 2022).

2.4: Predictors of interest

Demographic data were extracted from the EHR, and included race, ethnicity, sex, and primary language spoken. Race is reported in descriptive statistics and in univariate analysis, but was dichotomized into White/non-White in multivariate analyses to preserve statistical power. Information on service use was based on CPT codes submitted by clinicians, and diagnoses were based on the ICD-9 or ICD-10 code at the last recorded visit (see online supplement for list of codes). Diagnoses were classified into five groups: schizophrenia/schizoaffective disorder, other unspecified psychosis, substance induced psychosis, primary affective disorders, or other non-primary psychotic disorders. Cannabis use at entrance and at last contact with the CSC program as well as lifetime use was manually extracted from the records based on clinical notes. Use on entrance to the CSC program was assessed in the initial evaluation note. Use at last contact with CSC was defined as any cannabis use noted in the final clinical encounter before exit. Lifetime use was defined as any mention of cannabis use in the medical record. Presence of alcohol use was ascertained by presence of this diagnosis in ICD-9 and 10 codes in the year prior to ending CSC. Medication adherence was tracked by number of months with a documented prescription of an antipsychotic medication in the 12 months before discharge, including oral and long-acting injectable antipsychotics of all classes. Relapse prior to ending CSC program services was operationalized as any ED visit or hospitalization for a psychiatric reason 12 months prior to ending CSC services. Any ED visit or hospitalization prior to enrollment in CSC was excluded. Baseline functional indictors (education, employment status, and living status) and tobacco use were extracted from initial evaluation notes.

2.5: Statistical methods

We calculated descriptive statistics of all demographic and clinical variables. To investigate predictors of relapse in the twelve months after discharge, logistic regression models were estimated for the main predictors of interest. Odds ratios (OR) were used with 95% confidence intervals. Multivariable models included age, sex, race, language, insurance, duration of treatment in the FEP program, medication prescriptions, cannabis use, comorbid alcohol use, and primary diagnosis. The same multivariable regression model was estimated to predict engagement in psychiatric care after CSC discharge. Predictive margins methods were used to generate estimates of outcomes after adjustment for confounders (Graubard and Korn, 1999).

Three sensitivity analyses were run. First. to investigate the change in program structure over time, analyses were run with all patients and then only with patients who received some services after the addition of peer support and supported employment and education services in 2018. Secondly, to check the robustness of the effect of substance use on outcomes, analyses were run with and without individuals with a primary diagnosis of substance induced psychosis. Finally, to account for variables that were reliably collected at intake but not at follow-up, an additional model was run including baseline substance use and functional variables.

Results were considered statistically significant at p=0.05 (two tailed). All analyses were conducted in Stata version 16.1.

3. Results

3.1. Clinical and demographic characteristics

The sample was predominantly reported as male sex (n=110, 77%). Race and ethnicity on record were as follows: sixty-eight (47%) White, 28 (20%) Black, 20 (14%) Hispanic, and 18 (13%) as Asian or South Asian. Twenty-two (15%) spoke a language other than English as their first language. At time of discharge 85 (59%) had a diagnosis of schizophrenia or schizoaffective disorder, 23 (16%) with an unspecified psychotic disorder, 5 (3.5%) with substance induced psychosis, and the remainder with mood disorders or other non-primary psychotic illnesses. Forty-six (34%) reported active cannabis use at last contact with CSC. Other demographic and clinical variables can be found in Table 1.

Table 1:

Demographics of cohort at time of last CSC contact (n=143)

Variables	mean	SD
CSC treatment duration (days)	539.7	415.9
Age at discharge	23.0	4.1
Months with antipsychotic fill 12 months predischarge	5.3	3.9
Variables	N	%
Race
White	68	47.6
Black	28	19.6
Hispanic	20	14.0
Asian or South Asian	18	12.6
Other or unknown	9	6.3
Sex
Female	33	23.1
Male	110	76.9
Language
English	121	84.6
Non-English	22	15.4
Insurance
Private insurance	85	59.4
Medicaid	48	33.6
Medicare	4	2.8
Self-pay	6	4.2
Education level at entry
Less than high school	35	24.48
High school graduate or equivalent	38	26.57
Some college	53	37.06
College graduate	17	11.89
Employed or in school at entry
No	67	46.9
Yes	76	53.2
Housing at entry
Living with family	104	72.7
Living independently	32	22.4
Other (ex. group home, homeless)	7	4.90
Diagnosis at discharge
Schizophrenia or schizoaffective disorder	85	59.4
Other psychosis NOS	23	16.1
Substance induced psychosis	5	3.5
Affective disorders	20	14.0
Other non-psychotic disorder	10	7.0
Tobacco use at entry
No use	104	75.4
Any use	34	24.6
Cannabis use at intake
No use	94	66.7
Any use	47	33.3
Cannabis use at discharge
No cannabis at discharge	90	66.2
Any cannabis at discharge	46	33.8
Alcohol use at exit
No alcohol use	128	89.5
Alcohol use	15	10.5

3.2. Predictors of relapse

In unadjusted comparisons, there were no differences in post-CSC relapse by race, language, sex, age, insurance, months enrolled in CSC, cannabis or tobacco use at intake, diagnosis, functional status at intake including education, employment, or living status, before ending CSC. In raw analyses, number of antipsychotic prescriptions was associated with increased relapse after discharge, but not in adjusted analyses (supplemental table 1). In both unadjusted and adjusted analyses, any cannabis use at last contact with CSC (aOR=4.06, p=0.004) and any ED visit or hospitalization 12 months prior to ending CSC services (aOR=4.69, p=0.002) were associated with a higher risk of relapse compared to the whole sample (Table 2). The predicted probability of relapse was 42.2% in individuals using cannabis at exit compared to 18.5% in those not using, a difference of 23.8% (95% CI 7.7–39.8%). The probability of relapse was 40.0% in those with a prior ED visit or hospitalization during CSC and 14.4% in this without, a difference of 25.6% (95% CI 10.3–40.9%).

Table 2:

Predictors of relapse in 12 months after discharge

	OR	p-value	95% CI
Any relapse 12 months before ending CSC	4.69	0.002	1.78	12.34
Months in CSC	1.004	0.87	0.96	1.05
Non-White race	0.89	0.80	0.36	2.22
Male sex	0.53	0.23	0.18	1.52
Non-English speaking	1.32	0.69	0.34	5.18
#months with antipsychotic fill	1.07	0.33	0.93	1.24
Age at exit	0.96	0.45	0.85	1.08
Cannabis use at last CSC contact	4.06	0.004	1.56	10.56
Schizophrenia or schizoaffective disorder diagnosis	1.01	0.99	0.35	2.9
Alcohol use at exit	1.491	0.59	0.35	6.34
Insurance
Medicaid	0.71	0.50	0.27	1.92
Medicare	0.81	0.88	0.055	12.07
Private Insurance (comparator)	1	.	.	.
No insurance	2.01	0.53	0.22	18.05

3.3. Engagement in follow-up care

In unadjusted comparisons, there were no significant differences by race, sex, age, insurance, diagnosis, tobacco use, functional status at intake, prior relapse or language in the odds of seeing a psychiatrist or therapist 12 months after CSC. More months of antipsychotic prescription before exit predicted psychiatrist follow up in raw but not adjusted analyses. (supplemental table 1). However, any cannabis use at exit from CSC was associated with lower odds of seeing a psychiatrist after leaving CSC in both unadjusted and adjusted analyses (aOR=0.32, p=0.03). The predicted probability of seeing a psychiatrist after exit was 32.8% in those who did not use cannabis and 14.9% in those who did, a difference of 17.9% (95% CI 3.6–32.1%). There was no significant association of cannabis on seeing a therapist after exit from CSC. CSC duration in months had a significant but small positive association with seeing a psychiatrist (aOR=1.04, p=0.04) and therapist (aOR=1.05, p=0.04) after ending CSC services. (Table 3).

Table 3:

Predictors of any engagement with psychiatric follow-up 12 months after discharge

	Any MD visit				Any therapist visit
	OR	p-value	95% CI		OR	p-value	95% CI
Any relapse 12 months before ending CSC	0.82	0.66	0.34	1.98	1.14	0.786	0.43	3.02
Months in CSC	1.04	0.04	1.002	1.08	1.05	0.04	1.003	1.09
Non-white race	0.64	0.31	0.27	1.53	2.23	0.11	0.85	5.85
Male sex	0.76	0.60	0.28	2.11	0.42	0.12	0.14	1.24
Non-English speaking	0.66	0.56	0.16	2.72	0.36	0.23	0.067	1.91
#months with antipsychotic fill	1.064	0.32	0.94	1.2	1.04	0.58	0.91	1.18
Age at discharge	1.012	0.83	0.91	1.13	0.93	0.25	0.82	1.05
Cannabis use at last CSC contact	0.32	0.03	0.12	.90	0.53	0.26	0.17	1.6
Schizophrenia diagnosis	0.62	0.37	0.21	1.77	0.52	0.26	0.17	1.64
Alcohol use at exit	1.14	0.86	0.27	4.85	0.62	0.59	0.11	3.54
Insurance
Medicaid	0.54	0.21	0.21	1.40	0.44	0.16	0.14	1.37
Medicare	0.75	0.82	0.064	8.89	4.93	0.18	0.48	50.30
Private Insurance (comparator)	1	.	.	.	1	.	.	.
No insurance	*	.	.	.	0.97	0.98	0.09	10.68

^*Predicted failure perfectly due to small n (n=6), could not interpret HR

3.4. Sensitivity analyses

To ascertain whether there was any difference in outcomes before vs after full implementation of the CSC model, the multivariate model was run both with and without patients who ended services before full CSC staffing was available. Thirteen participants were excluded because they left the CSC before 2018 when peer support and supported education and employment services were available. There were no changes in the relapse findings. For psychiatrist follow-up, the point estimate was similar but with a smaller n significance was at trend level (OR=0.37, p=0.063).

To ensure that the substance use findings were robust, the multivariate model was run with and without individuals with substance-induced psychosis. Five participants were excluded because of a diagnosis of substance-induced psychosis, and there were no changes in any results.

Finally, as a sensitivity analysis incorporating predictors only available at intake, an additional model was run adding baseline tobacco and cannabis use as well as baseline functional predictors. Results were unchanged for the main analysis of relapse risk after exiting CSC. No engagement in employment or school at intake was associated with significantly lower odds of psychiatrist (OR=0.27, p=0.021) and therapist (OR=0.24, p=0.029) follow-up, and cannabis use at exit was associated with lower follow-up with a similar point estimate but only trend level significance (OR=0.31, p=0.058). All sensitivity analyses are available in the online appendix.

4. Discussion

Individuals who had a relapse after leaving CSC services were more likely to have had a relapse while participating in the CSC and to have used cannabis at time of last CSC contact. Cannabis use was also associated with a lower odds of seeing a psychiatrist after CSC, and duration of time enrolled in CSC had a small positive association with psychiatrist and therapist follow-up after CSC. Given both higher rates of relapse and lower rates of follow-up in individuals using cannabis, this provides evidence that this group merits both targeted outreach after ending services as well as novel program development to assist these individuals in continuing their recovery. In this urban safety net sample, there were also no evident differences in relapse or follow-up by race or ethnicity, which contrasts with other analyses finding higher rates of post-intervention relapse among minoritized racial and ethnic groups (Humensky et al., 2021).

Findings that ongoing cannabis use at the end of the CSC program is associated with increased relapse risk is consistent with research from larger population-based studies of people with psychosis (Schoeler et al., 2016a). One study found a stronger relationship with more frequent and higher potency use (Quattrone et al., 2021), whereas another study of RAISE data showed a higher risk for sporadic cannabis users rather than heavy users (Wright et al., 2022). RAISE data showed that engagement in CSC had a negligible effect on cannabis use (Cather et al., 2018). This is of concern given that we have demonstrated that patients who continue to use cannabis when leaving CSC represent a group at high risk for relapse and may merit more intensive interventions.

Our secondary analysis suggested that individuals who used cannabis were also less likely to follow up with a psychiatrist after CSC. While disengaging with first episode psychosis services may not necessarily represent a bad outcome (ex. an individual recovering and dropping out because they perceive no need for further support), the combination of less follow-up with higher relapse rates among individuals using cannabis indicates this is not likely a benign phenomenon in this population. Prior research has demonstrated a higher rate of dropout from first episode psychosis services among individuals using cannabis (Mascayano et al., 2020; Miller et al., 2009), and this study suggests this extends into the year after exit from early intervention services. Of note, there was no significant effect of cannabis on therapist follow-up, which could be simply be due to limited statistical power or other clinical factors, such as cannabis use leading to antipsychotic discontinuation and dropping out of care with a psychiatrist while continuing with therapy. It is also important to note that follow-up was low overall (only 32.8% even in the non-cannabis group), indicating there is still great unmet need for policies that can improve care engagement after CSC discharge.

While CSC duration and medication adherence did not affect relapse risk, there was a small positive association between months enrolled in CSC and increased follow-up after ending CSC services. This could represent a beneficial effect on treatment engagement leading to ongoing willingness to accept care, but could also represent a subgroup of individuals with a worse clinical course who are held in CSC services longer due to staff, family, or patient concerns. Also of note is the lack of effect of antipsychotic prescription, which could be explained by diagnostic heterogeneity of the sample as well as a limited 1-year follow-up, which may not have been long enough to fully elucidate this. Further, this measure was not able to take into account the timing of the prescription (ex. several months before exit vs the months leading up to exit). This could be explored further in future research with more prospective antipsychotic adherence tracking.

There have been multiple negative trials of specialized interventions to reduce cannabis use in FEP (Bonsack et al., 2011; Madigan et al., 2013), and at this time no high-quality evidence exists to address cannabis use in psychosis. Qualitative research indicates that changes in cannabis use are usually driven by changes in social or role functioning and less by concerns about psychotic symptoms (Marino et al., 2022), suggesting that integrating substance use treatment into CSC within existing peer or employment support with a focus on functional outcomes may have more success. CSC programs should also establish closer linkages with dedicated dual diagnosis programs in their area for step down care after CSC, and states could expand training and technical assistance to community programs to serve this population. Certified Community Behavioral Health Center (CCBHC) models have shown early success in increasing availability of specialty mental health and substance use care under one roof (Wishon and Brown, 2023), and the expansion of this model could provide an important avenue for follow-up care for individuals leaving CSC with ongoing substance use.

The other significant predictor of relapse after leaving CSC was the presence of any relapse in the year before exiting the CSC program, consistent with other some international studies (Kam et al., 2015) but not others (Puntis et al., 2018). This group is more likely to have had at least two significant psychotic episodes, one necessitating referral to the CSC and one while in the CSC, and prior research has shown that with two or more episodes the risk of further episodes and functional decline is high (Lin et al., 2021). This represents a group of patients who would benefit from engagement in specialized psychosis programs or more intensive services like PACT after CSC given the likelihood that this group will need more intensive longer-term support. Further research should investigate trajectories of service use after CSC and their effect on longer term outcomes.

The present study did not show a difference in relapse risk by race in either unadjusted or adjusted analyses. Prior research has suggested that the components of CSC and structured approach to treatment result in equitable outcomes across racial groups for those seeking treatment (Chin et al., 2012). The present study demonstrated that relapse rates were similar across race and ethnicity after leaving CSC, contrasting with the six-month outcome study after discharge showing higher rates of hospitalization for Black and Hispanic patients before and after CSC services ended (Humensky et al., 2021). However, for reasons of statistical power, in multivariate analyses race and ethnicity were collapsed to binary categories of white and non-White, which limited our ability to ascertain unique trajectories by race or ethnicity and assumes that all non-White individuals are more similar to the white population, which may or may not be true. Additionally, racial disparities in mental health care often result from differential rates of treatment initiation rather than differences for people in care (Cook et al., 2017), and this present study did not address the factors that explain who enrolls into CSC. Further research should seek to replicate these findings in larger community samples as well as in more diverse service environments to ascertain longer term trends in outcome.

The present study has several limitations. First, it is a retrospective observational analysis design which limits the causal interpretation of results. On the other hand, our case mix and development of services over time likely mirrors the real-world operations of many CSC clinics, which increases generalizability. Second, while we had some ability to follow participants lost to follow up through claims data, these data were incomplete and did not allow for direct comparison of relapse rates before and after CSC. This also limits our ability to investigate those lost to follow up care at our institution. Third, data were derived from the medical record as opposed to prospective structured assessments. This did allow us to capture individuals who would be less likely to complete standardized assessments, but meant that the results were limited to what was reported in the EHR. For instance, detailed descriptions of frequency or intensity of cannabis use before FEP were generally not available, and other established risk factors like duration of untreated psychosis (DUP) were not consistently reported limiting our ability to examine the intersection of DUP and other predictors. Fourth, for our medication variables, claims data were not fully available meaning we could not reliably ascertain whether individuals were filling their antipsychotics, and we were not able to differentiate clozapine and long-acting injectable antipsychotics from other antipsychotic medications. Given their known benefit for relapse prevention, future studies could investigate how different medication regimes moderate or mediate our findings. Finally, although we analyzed possible differences in pathways from referral-to-participation in the CSC study cohort and found no significant differences by race/ethnicity and language, this real-world secondary data analysis could not account for possible differences in patients’ likelihood of receiving referral into the FEP specialty program.

Future studies could expand on these findings by including multiple sites and larger sample sizes through multi-site collaborations to increase power to investigate some of the associations found, and more complete claims data would also allow more direct evaluation of the longevity of treatment effects in CSC. Qualitative research could also incorporate the perspectives of people accessing services to ascertain the factors underlying their trajectories.

5. Conclusions

Relapse prior to ending CSC services was associated with relapse after leaving CSC. Cannabis use was associated with both a higher rate of relapse and lower rates of seeing a psychiatrist in the year after ending CSC. White and non-white patients had similar rates of relapse and follow up in the year after leaving CSC. Given ongoing interest from clinicians and policymakers in transitions and step down from CSC (Daley et al., 2022), these results can inform risk stratification and discharge planning with a particular focus on individuals more likely to have poor outcomes after leaving specialized early psychosis care.