Premature termination, defined as when a “client has left therapy before obtaining a requisite level of improvement or completing therapy goals” (Hatchett and Park, 2003, p. 226) is a significant and widespread problem in the field of mental health (Barrett et al., 2008). Up to 50% of clients discontinue psychological services prematurely (Barrett et al., 2008) which undermines the potential benefits of treatment and reduces the cost-effectiveness of these interventions (Westmacott et al., 2010). Several studies examining potential variables associated with this phenomenon (i.e., client, therapist, and treatment) have been conducted (Wierzbicki and Pekarik, 1993; Garfield, 1994; Sales, 2003; Clarkin and Levy, 2004), but results are largely inconsistent due to the plethora of terms that are often interchangeably used (i.e., attrition, dropout, early termination, pre-mature termination, early withdrawal, among others) (Wierzbicki and Pekarik, 1993) and the methods adopted to operationalize these constructs (Swift et al., 2009). To illustrate, main categorizations of dropout usually include duration of the therapy (i.e., when the YP in a study terminates treatment before the pre-defined cut off) and therapist judgment of whether the treatment termination is a dropout. However, it is often difficult for therapists to detect how clients are responding to therapy (Hannan et al., 2005). Nonetheless, therapists' and the YPs' or carers' assumptions about treatment goals and expectations may differ (Barrett et al., 2008), leading to non-mutually agreed decisions (de Haan et al., 2013). In fact, whether or not criteria for “clinical improvement” or recovery have been met, clients may prematurely end treatment because the necessary gains in functioning have been obtained prior to the end of a set number of sessions, or because they may want to try other interventions on their own, outside of treatment.

On the other hand, clients may recognize a lack of improvement and believe that additional sessions will not be helpful, another perspective that can also be difficult to detect during therapy (Lambert et al., 2005). Further, the type of treatment a client receives also influences rates of non-mutually agreed endings in therapy (Barrett et al., 2008). Treatments involving both medications and therapy in the extant literature have consistently shown lower rates of attrition than either medication or therapy alone (Arnow et al., 2007). Another reason for dropout includes the YP's diagnosis (Westmacott et al., 2010). Researchers reported higher rates of attrition among clients with more severe diagnoses (i.e., externalizing problems) and more complex diagnostic pictures (i.e., comorbidity) (Thormählen et al., 2003). There is also some evidence showing that external factors may also influence YP's use of health care services or constitute barriers to continuing treatment. Such factors include difficulties in finding mental health services, cost for services, degree of family involvement, and social support networks. Beyond that, practical issues such as greater distance traveled, scheduling conflicts, and long waiting lists can negatively influence community perception of the mental health services resulting in earlier dropout from care (Westmacott et al., 2010). Therefore, a need-based definition is a valuable method for categorizing treatment dropouts and mitigates disadvantages of existing definitions of dropout (Dossett and Reid, 2019).

Demographic data such as belonging to an ethnic minority (de Haan et al., 2018) or lower socioeconomic status group (de Haan et al., 2014), having a younger mother, and living in a single-parent household (de Haan et al., 2013) are social and family variables that increase the likelihood of dropping out of treatment. Despite this, variables related to the treatment itself and those related to the therapist were also found to be overall stronger dropout predictors than the pre-treatment child and family or parent/carer variables. Specifically, dropout increases when adolescents experience lower quality of the therapeutic relationship, lower perceived relevance of treatment, more treatment participation barriers, and more stressors (Carter, 1995; Garcia and Weisz, 2002). Significant predictors of dropout are also the adolescent's experience of their therapist as being directive, controlling, and confronting; the therapist not showing care and concern; and dissatisfaction with the focus of therapy (Jethwa et al., 2019). More cancellations or no-shows over the course of the treatment have also been consistently found as a reliable predictor of dropout (Kazdin et al., 1994; Chasson et al., 2008; de Haan et al., 2013). Emerging evidence highlights the importance of including cultural understanding and adoption in the therapeutic relationship in order to retain YP in mental health settings (Yeh et al., 1994; Carter, 1995; Cunningham et al., 2002; Lau, 2006; Huey and Polo, 2008; Miller et al., 2008; Bibi et al., 2017).

Nonetheless, treatment dropout is often regarded as a negative outcome in therapy. A mixed-method study that explored YP's reasons for dropout from treatment highlighted that nearly one third of the sample indicated they had received satisfactory treatment or experienced symptom improvement (O'Keeffe et al., 2019a). Therefore, when clients end treatment non-mutually, their therapists are often not aware whether their clients were (dis)satisfied with the therapy (Westmacott et al., 2010). However, therapists of YP who dropped out due to symptom improvement reported they were not clinically concerned about this group of dropouts. This indicates that treatment terminations following clients who benefitted from therapy may not yet meaningfully be accounted for in existing explanatory models of dropout from treatment (O'Keeffe et al., 2019a).

There is a growing interest in improving outcomes for YP accessing mental health services, with the main focus thus far being on improving symptoms and aiming for “recovery”. However, other outcomes may also be important, and in particular, whether YP and therapist mutually agree with the end of treatment. This line of reasoning may have implications for the evaluation of outcomes at the case level and service level, including accuracy of data and effective use of costs to the National Health Services (Mental Health Taskforce to the NHS in England, 2016). Therefore, it is important to broaden our understanding of the influence of demographics, referral process, and symptom improvements on case closure. Evidence that non mutual case closure may not necessarily be a problem may reflect more self-efficacy, competence, self-rated improvement, and autonomy among YP and their carers (Simon et al., 2012; O'Keeffe et al., 2019b). Despite this wealth of knowledge, we are yet to fully understand if symptom improvement is an indicator of a “good” outcome.

In this vein, the present study aimed to examine whether levels of meaningful improvement in symptoms were associated with reasons for ending treatment, using multilevel multinomial regression analysis controlling for age, sex, ethnicity, and referral source. We hypothesized that youths whose problems meaningfully improved were more likely to mutually agree to end treatment.

Three datasets held by the Child Outcomes Research Consortium on children and young people (0-25 years old) who accessed mental health services in the United Kingdom (UK) between 2002 and 2019 were merged (Costa da Silva et al., submitted). The data corpus was collected by clinicians and service administrators from YP mental health services across England, including those participating in a programme offered by the National Health Services to implement evidence-based practice between 2011 and 2015 (Fonagy et al., 2017). From this merged dataset, cases were included in the present analysis if: (a) the child or young person was aged 6-25 years to reflect the age range that the included measures could be self-reported, (b) the case was closed, (c) there was at least one paired outcome measure completed at time 1 and time 2, and (d) there was a reason for case closure. This resulted in a final dataset of N = 8,995 episodes of care (i.e., independent observations) [Female = 5,469, 61%; meanAge = 13.66 (SD = 2.87) years]. Detailed demographic characteristics are shown in Table 1.

The present analysis involved secondary analysis of anonymised administrative data and therefore, an ethical review was not required (Tripathy, 2013).

To examine whether YP who meaningfully improved were more likely to mutually agree to end treatment, accounting for the nesting of episodes of care in services and controlling for age, gender, ethnicity, and referral source, multilevel multinomial logistic regressions were conducted in STATA 16 (StataCorp., 2019). Three preparatory models were estimated. In Model 0 (null model) the variance explained in case closure reason at the service-level was examined and no predictors were added. The intraclass correlation coefficient was 45%, indicating that there was significant service-level variation in case closure reason and confirming that multilevel modeling was the appropriate statistical approach. In Model 1, demographic characteristics were added: male; grand-mean-centered age; and ethnicity with the White British group as the reference category as it was the largest group. In Model 2, referral source was added with primary care as the reference category. In the final model, meaningful change was added with no change selected as the reference category as it was the largest group. The likelihood ratio test was used to compare successive models, which were significant, and all variables were therefore retained in the final model. In particular, the likelihood ratio test was significant for the final model compared to Model 2: χ²(6) = 111.3, p < 0.001.

Although our findings suggest that YP who meaningfully improve are more likely to mutually agree to ending treatment, clinicians and researchers should consider that some YP may non-mutually end treatment if they self-assess as having sufficiently improved. This speaks in favor of ongoing evaluations of treatment goals and progress tracking. In light of the previous literature, it is also important to note that families with YP diagnosed with specific difficulties, having additional complexities, or experiencing external variables such as deprivation are more vulnerable to non-mutually end treatment. Therefore, researchers, clinicians, families, YP, and decision-makers should continue to work together to develop tailored service level programmes and individual interventions to ensure underrepresented and underserved families are reached. For example, the finding that YP from non-White British ethnic groups are more likely to drop out of treatment highlights the importance of reaching these groups. This includes considering the referral routes and types of interventions offered, including consideration of community-based interventions, which may widen reach and increase retention for the identified groups.

One area that was not possible to investigate in the present study is the parent/carer perspective, given the significant role parents and carers have in YP retention in mental health settings (Weisz et al., 1987; Garcia and Weisz, 2002). Future research should explore this, as parent/carer views may differ from those of the clinician and the young person. It is also important to continue research into the use of digital interventions. With growing interest in this area, through web-based appointment systems and texting to mobile phones, we may be able to better capture reasons for treatment dropouts and facilitate non-face-to-face support for YP. Further qualitative and quantitative studies are also welcomed to explore YP's own descriptions of good outcomes and treatment ending to triangulate or further develop our current descriptions.

A major strength of this study is the inclusion of a large sample size. Moreover, we investigated the factors associated with case closure and YP's mental health using multilevel modeling, a method that was able to account for of individual and service-level variation. However, these results should be interpreted in the context of several limitations. The large majority of the participants were female and identified as White-British ethnicity thus preventing us from making predictions on the impact that cultural variations may have on the study's findings. Moreover, the specific problems presented by YP may have influenced the study outcome, but we were unable to account for this in the present investigation.

Another limitation of the study relates to the numerous ways that dropout can be defined, bringing challenges to the ability to compare results between studies (Barrett et al., 2008; de Haan et al., 2013). Reliability of the study's results is also affected by the absence of detailed information on professionals' reasons for case closure and the lack of qualitative data from YP or parents/carers in order to provide a deeper understanding of the current sample. In addition, the unavailability of follow-up data prevents drawing conclusions about the efficacy and effectiveness of the intervention – therefore on the extent to which clients' decision to discontinue the therapy due to perceived improvements or dissatisfaction is supported by trends in symptoms or clinical outcomes. Without a randomized controlled design, inferences about causation, of symptom improvement and reason for case closure, cannot be made. Another constraint identified was the reliance on routine pre-collected data, resulting in less flexibility to include explanatory variables of interest, such as the parent/carer perspective. Although this may compromise the rigorous empirical research standards and cause-effect relationships, this method has the benefit of allowing us to investigate variables without additional research participation burden to YP (Mansfield et al., 2020).