Disengagement From Early Intervention Services for Psychosis: A Systematic Review

We conducted a search of scientific publications in the following databases: PubMed, Google Scholar, EBSCO, Ovid, and Embase. A wide range of keywords in English, Spanish, Portuguese, and French were included. We kept search terms broad to find relevant studies, even when the specific keyword, “disengagement,” was not mentioned in the title or abstract. Given the lack of consensus in terminology as noted above, we were interested in any form of disengagement. Accordingly, we included terms such as “disengagement,” “engagement,” “nonadherence,” “dropout,” and “discharge” in combination with “psychosis,” “early intervention services,” “first-episode psychosis,” “psychotic disorder,” and “schizophrenia.” We also manually checked the references of the eligible reports. We searched for articles published from January 1966 to June 2019, including the time frame (2000–2012) previously used by Doyle et al. (6). The terms were also combined for a more precise search and used to identify titles, abstracts, and full texts in the databases noted above.

Articles were included in this review if they reported rates of disengagement (or similar concepts, such as engagement, nonadherence, no therapeutic alliance, and dropout) from EIS; explored predictors of disengagement considered here as predisposing, enabling, and need factors; reported primary data published in peer-reviewed journals; included quantitative or mixed methods; and were written in English, Spanish, Portuguese, or French. We reviewed each report through a sequential process (searching, refining search strategy, examining titles and abstracts, and reviewing full articles). Two reviewers (E.v.d.V., G.M.-A.) independently rated the methodological quality of each report on the OCCS scale. The two reviewers met at the beginning, midpoint, and final stages of the review process to discuss challenges and uncertainties related to study selection. If any disagreements regarding study inclusion arose, the reviewers reached consensus on the final decision.

Three researchers (F.M., E.v.d.V, G.M.-A.) collectively developed a data-charting form and determined which variables needed to be included to answer the research questions. This form was based on the framework set out by Doyle and colleagues (6) and included location of the study, sample and research setting, definition of disengagement, disengagement rates and time, and predictors of disengagement.

We used the OCCS scale to evaluate the methodological characteristics of the studies (11). This instrument assesses methodological quality across 14 dimensions (research question, study population, groups recruited from the same population and uniform eligibility criteria, sample size justification, exposure assessed before outcome measurement, sufficient time frame to observe an effect, different levels of the exposure of interest, exposure measures and assessment, repeated-exposure assessment, outcome measures, blinding of outcome assessors, follow-up rate, and statistical analyses), and has been used in previous systematic reviews of cohort and cross-sectional studies (12, 13). The reviewers of study quality (F.M., E.v.d.V, G.M.-A.) could select “yes,” “no,” or “cannot determine/not reported/not applicable” in response to each item on the quality assessment tool.

On the basis of the sum of items on the quality assessment tool, each study received a high-, medium-, or low-quality rating. Generally, a study rated as high quality had the least risk of bias, and its results were considered to be valid. A study rated as medium quality was susceptible to some bias deemed not sufficient to invalidate its results, and a low rating indicated significant risk of bias.

After the two independent raters (E.v.d.V. and G.M.-A.) rated each selected paper, a third independent rater (A.R.H.) calculated the kappa interobserver agreement for the National Institutes of Health tool (intraclass correlation=0.7) in Stata 12. All three raters then had a series of meetings to resolve disagreements by rereading the studies and further discussing the studies’ methodological features, results, and implications.

Study populations varied notably in terms of their demographic features, with sample sizes ranging between 21 and 786 participants, ages between 14 and 64 years, and the proportion of females between 22% and 53%. All studies featured individuals with first- or second-episode psychosis attending a variety of EIS programs, but the studies slightly differed in inclusion criteria, with one study not reporting any inclusion criteria. Studies were conducted in North America (N=7), Europe (N=4), Asia (N=4), Australia (N=3), and New Zealand (N=2); no studies were from Africa or South America (Table 1). The length of follow-up ranged from 9 to 36 months, and half of the studies (N=10) measured disengagement within 24 months of follow-up. Most studies (N=19) did not report sample size calculations and did not include multiple time-point measurements of independent time-varying predictors.

The EIS disengagement rates reported in each study are shown in Table 1, ranging widely from 12% (14) to 53% (15) within 36 and 24 months of program initiation, respectively. There was no consensus definition of disengagement across studies. Definitions varied broadly from “termination of treatment despite therapeutic need” (16) to no contact with clinical team “after all possible ways to engage had been explored over a 2- to 3-month period” (14). One study defined three different types of disengagement: continuous default from service despite therapeutic need and active tracing by staff (type I), continuous default from service and reengagement through hospitalization (type II), and continuous defaults from outpatient appointments and reengagement through outpatient clinics (type III) (17).

Nonetheless, we could identify some common features across these definitions. First, most studies considered “no contact” with and “active refusal” to clinical team or staff as a key outcome associated with leaving treatment (7, 8, 25). Moreover, several studies operationalized disengagement as dropping out from treatment “despite clinical advice or judgment” (19, 26), which suggests that stopping services against clinicians’ recommendations is usually seen as a negative outcome. Treatment disengagement was usually reported or rated only by clinicians, which substantially limits our understanding of why clients may believe that services are no longer required. Second, just two studies stated that disengagement occurs “after all possible ways to engage clients” took place (14, 20). Only one study described exactly what kinds of strategies were used to reengage clients: “these included appointment letters, phone calls, text messages, e-mails, home visits and contact with family, friends and other health, education and social care providers” (14). Third, the highest disengagement rates generally corresponded to definitions of disengagement that included concepts of “dropping out” (7), “treatment retention” (21), “terminated treatment against clinical advice” (26), and “lost to any contact with mental health services” (27).

In addition to varying in disengagement definitions, the studies differed in how long participants were followed up. The follow-up assessments ranged from 9 to >36 months after EIS initiation. The lowest rates were associated with definitions considering “all possible ways to engage them had been explored” (14) and “active tracing from staff for psychiatric follow-up” (28). Some studies used scales to measure levels of engagement or disengagement, such as the Service Engagement Scale (29, 30) and the Singh O’Brien Level of Engagement Scale (31). Finally, the sole randomized controlled trial identified in our search reported much lower rates of disengagement among those participating in the experimental EIS condition (32%) than among those receiving usual care (76%) (24).

Table 1 also summarizes findings related to predictors of disengagement (14–29). We reorganized these predictors as predisposing, enabling, and need factors on the basis of Andersen’s model (Figure 1). For the purpose of organizing significant predictors according to Andersen’s model, we selected only studies with a high quality rating (N=13).

A few enabling and need factors were consistently associated with EIS disengagement, including lack of family support, lower medication adherence, and drug misuse. Some studies also found a relationship between illness characteristics at baseline and later disengagement, such as fewer negative symptoms and higher functioning (7, 14, 17, 20). Moreover, several predisposing factors, such as older age, male sex, and belonging to a racial-ethnic minority group, were often reported; however, although some studies found associations of these factors with disengagement from EIS, most did not. In addition, the relationship between some of the factors and disengagement was inconsistent across studies. For instance, depending on the study, both longer and shorter durations of untreated psychosis were associated with an increased likelihood of disengagement. Overall, our findings highlight the heterogeneity of the relationship between measured characteristics and disengagement but provide evidence that factors associated with the level of support, adherence to a medication regimen, and comorbid conditions were most predictive of disengagement.

One reason for the substantial between-study heterogeneity in the identified predictive factors may be variation in the conceptualization and measurement of disengagement. This has been also noted in a recent review of definitions of disengagement, which found that only a few studies used a similar definition (i.e., when reports came from the same research group), but most reports defined disengagement as a dichotomous, binary variable (engaged versus disengaged) (32). In our review, we identified three common features across studies when they referred to disengagement: “no contact” with and “active refusal” to the clinical team or staff, leaving services “despite clinical advice,” and failure to reengage clients after several attempts to do so. Although some people may leave treatment prematurely because of negative experiences with the care received or because little in their lives changed as a result of that care, others may stop treatment because they feel better and decide that services are no longer required, even though the clinical team might think otherwise. They may leave for circumstantial reasons, such as accidents or moving away from the area served. These heterogeneous reasons for discharge remain largely unexplored, despite their potential importance for predicting postdischarge outcomes (e.g., relapses) and care transitions. Some reasons for terminating EIS reflect, in fact, disengagement, whereas others may refer to processes to and through care that represent affirmation of care and treatment benefits (e.g., positive discharge).

Previous research has linked family support to engagement in mental health care among individuals with early psychosis. For instance, Lucksted et al. (33) found that family members played a crucial role in promoting engagement among individuals with early psychosis by reminding them about clinical appointments, encouraging attendance, and providing transportation. However, family involvement does not necessarily imply “family support.” In other words, some families could also present barriers to clients’ participation in treatment by arguing against providers’ recommendations and by interfering in participants’ choices (33). EIS staff should listen to family members, provide support and compelling information about psychosis, and help participants form their own opinions and navigate family relationships and boundaries. Dialogic practices could be of use to better understand family dynamics and crises, discuss issues from different perspectives, and implement practical solutions that make sense to the client and family members (34).

Poor medication adherence has also been found to be related to leaving mental health services. Some authors have suggested that this finding is not surprising because it may reflect a participant’s general level of treatment adherence (35). Poor medication adherence might also represent low trust in the care that EIS offer or misconceptions about the clinical and therapeutic model used by service providers (35). Results from qualitative research have suggested frictions between the subjective meaning that clients attach to experiences of psychosis and the promotion of medication adherence from a biomedical perspective by some providers, which could lead to medication discontinuation and EIS disengagement (36). Therefore, a broader understanding of the pros and cons of taking medications, from both clients’ and providers’ perspectives, is needed. Additionally, medication adherence needs to be understood within a framework of shared decision making in which participants and EIS staff must find the balance between “the duty to care” (i.e., remaining engaged with clients no matter what decisions they make) and “the dignity of risk” (e.g., the right to make choices, to fail, and to learn) (3).

Finally, it has been widely reported that comorbid substance abuse is strongly associated with noninitiation of mental health care and disengagement from mental health services among individuals with psychosis. Substance abuse has several negative effects, such as increased relapses and hospitalization, impaired social and occupational functioning, and transinstitutionalization in jails and other non–mental health settings (3). Employing therapeutic strategies based on integrated dual-diagnosis treatment programs may help EIS staff to address comorbid substance abuse early in the course of treatment (3). These are evidence-based programs that emphasize outreach, comprehensiveness, and person-centered care. Under this model, clinicians use a broad set of techniques, including motivational interviewing and social and community-based interventions. Results from previous studies indicate that these programs improve treatment participation and enhance initial and ongoing engagement among those with dual diagnoses of mental illness and substance abuse (36, 37).

Given the steep rise of EIS in the past few years, here we had the opportunity to extend a previous review on the topic (6) by including a larger and more diverse set of studies. The use of a quality rating based on the OCCS scale enabled a more objective evaluation of the quality of the evidence and the development of methodological and data analysis recommendations for future research. Finally, our work sheds new light on the association between disengagement and clusters of baseline engagement factors by incorporating predictors of disengagement in a conceptual framework.

Some limitations of this review must also be noted. For instance, given the inconsistent findings across the studies examined, it is challenging to reach generalizable conclusions. At least part of the variance seen in this line of research might be explained by context-specific phenomena (e.g., health services organization). Furthermore, we could not conduct a meta-analysis because of the heterogeneity in research designs and definitions of the outcome. Also, selection bias may have been an issue across studies and may have influenced the performance of predictors.

On the basis of our findings, we propose a series of considerations that future research should take into account. First, given the inconsistency across studies regarding disengagement definitions, we suggest bringing key stakeholders together (e.g., through partnerships and community-partnered engagement with client and family advocacy groups) to reach a consensus, develop common metrics and measures of treatment disengagement, and design strategies to increase engagement when deemed reasonable. This discussion should include providers as well as patients and families, considering that they may have differing opinions and perspectives regarding participation in treatment and disengagement. For instance, Smith et al. (38) conducted qualitative interviews with 56 participants with psychosis who had disengaged from mental health care and with their service providers. Patients commonly expressed the following reasons for disengagement: services not relevant for their needs, lack of trust toward service providers, and the belief that they were not ill. Providers, in contrast, reported lack of insight, language and cultural barriers, and stigma as the main reasons. Such divergent perceptions support the importance of examining these issues from different perspectives. Furthermore, Cook et al. (39) recently reported that self-directed clients, compared with individuals in usual care, had better outcomes (e.g., recovery, education, and employment) but spent less time on recovery-oriented care, such as psychosocial rehabilitation, skills training, and case management. The motives behind these reduced service utilization trends are not yet well understood (5), and they may challenge what researchers and mental health professionals know about treatment length, participation in treatment, and disengagement.

Participatory methodologies, such as Delphi methods (40) and concept mapping (41), are useful to conduct such group activities. For instance, a group-model-building approach (41), which is a system dynamics–based method that has been used to guide the design, adaptation, and implementation of evidence-based practices in varying health settings, could be employed to engage stakeholders in identifying causal loop diagrams (including variables, relationships, and feedback) that represent the main factors affecting EIS disengagement in order to inform the development of improved engagement strategies. This approach would enable inclusion of a broader set of stakeholders, including individuals with early psychosis and their relatives, who are not usually represented in this kind of exercise and whose input is highly needed (42). Related issues might be also part of the discussion, such as perceived and evaluated need for services, unmet needs, satisfaction with services, optimal service length, reasons for staying in or leaving the program, discharge processes, and continuity of care.

Second, most reports corresponded to cohort studies that collected data retrospectively from clinical registries; just a few studies conducted their own assessments and quality assurance. Although data from medical records can be useful for programmatic planning and treatment supervision, its use for research requires a thorough analysis aimed at identifying and addressing multiple biases, such as confounding, selection bias, and measurement error. To move this line of research forward, future studies should examine causal relationships between potential predictors and disengagement while controlling for biases that may arise in data collection and analysis.

A third consideration is that to estimate causality between predictors and disengagement, prospective studies of how predictors vary over time are needed. Unfortunately, all the studies measured potential predictors of disengagement only at baseline, despite the fact that some of those factors change over time (e.g., medication adherence, symptoms, and social functioning). These time-varying variables could include confounders of the association between predictors of interest and disengagement, which might have a different impact on the outcome depending on the time point at which they are assessed. A series of statistical methods are now available to deal with time-varying variables, such as propensity scores and inverse probability weighting, that can be of use here (43).

Fourth, the selection of disengagement predictors in the studies reviewed here relied mostly on statistically significant associations between the predictors and the outcome in bivariate analyses. Generally, researchers used criteria for the identification and inclusion of predictors that were not based on predefined hypotheses or theories. Future research, therefore, should use well-defined conceptual frameworks to choose the variables to be analyzed and their relationships with each other and with the outcome.

Finally, we note that only one of the 13 high-quality studies assessed factors at the contextual level (18). It has been reported in other health settings that contextual factors such as poverty, accessibility and barriers to care, and characteristics of clinics and providers play an important role when engaging clients with services. These factors, which are best measured at the contextual level, are mostly unexplored in EIS contexts. Most EIS are built around therapeutic principles, such as person centeredness, shared decision making, and recovery orientation, yet little is known about how much those principles help to maintain engagement among clients.