RESEARCH DOSSIER
Engagement, Dropout, and Treatment Discontinuation
Abstract. Engagement with psychological treatment is an evolving relationship among patient goals, therapeutic processes, intervention demands, and health service conditions. This monograph critically synthesizes the history, definitions, methods, theoretical models, empirical findings, and unresolved problems concerning treatment initiation, attendance, premature termination, and discontinuation in psychotherapy and outpatient psychosocial care. It draws on original cross-national survey evidence, psychotherapy meta-analyses, intervention experiments, contemporary electronic health record prediction, patient and clinician perspectives, digital treatment research, and current clinical guidance. The literature establishes substantial but definition-sensitive attrition, pronounced early exit, associations of preference accommodation and therapeutic alliance with continuation, and evidence that preparatory role induction can reduce premature termination. Its strongest methodological conclusion is that administrative nonattendance, treatment dropout, and research follow-up loss measure different events; most observational predictors cannot establish causal mechanisms. A complete scientific account links referral opportunity, session-level experience, clinical outcomes, reasons for ending, service inequalities, and autonomous choice. The monograph proposes an evidence-sensitive program for multistate measurement, equitable intervention testing, transparent risk prediction, and safe continuity across human and digitally mediated care.
Keywords: psychotherapy engagement; treatment initiation; attrition; premature termination; psychotherapy dropout; treatment discontinuation; care retention; therapeutic alliance; treatment preference; shared decision-making; role induction; mental health service access; digital mental health; supportive accountability; clinical follow-up; missing data; survival analysis; health equity; patient autonomy; Artificial Era.
Introduction: The Scientific Problem of Continued Care
Engagement is the sequence of decisions and interactions through which a person enters, participates in, modifies, and eventually concludes psychological treatment. A treatment can be efficacious when delivered under controlled conditions yet exert little population-level benefit if eligible people cannot begin it or cannot sustain contact long enough to receive an appropriate therapeutic opportunity. The study of engagement therefore connects clinical psychology, counseling, service organization, behavioral science, and health equity. It also requires distinguishing the value of continued attendance from the value of treatment itself: appropriate care sometimes means changing clinicians, switching interventions, or ending an unhelpful encounter.
This monograph investigates the relationship among treatment uptake, sustained participation, missed appointments, discontinuation, and premature dropout in psychotherapy and associated outpatient psychosocial care. Its core questions concern how these events should be defined and measured, what predicts their occurrence, which mechanisms have credible empirical support, and what forms of intervention can make care more accessible, responsive, and safe. The central empirical difficulty is that nonattendance is an observable event while the clinical meaning of an ending is often latent. A missing appointment can mean symptom improvement, practical exclusion, disillusionment, transferred care, deterioration, or an autonomous decision to stop.
The large synthesis by Swift (2012) established both the substantial frequency of premature termination and the importance of operational definition. More recent evidence, including Giber (2026), extends the research into routine electronic health records and early risk prediction. The scientific task is to treat the whole care trajectory as the object of study while preserving differences between descriptive associations, tested interventions, and causal explanations.
Research Method and Evidentiary Scope
This work is a structured critical narrative synthesis. Sources were chosen to represent original population studies, observational clinical research, qualitative investigations, intervention trials and meta-analyses, methodological contributions, and current professional guidance. Bibliographic identity and reported central findings were checked against publisher, PubMed, journal, or official guidance records; accessible full texts were considered where available. This is neither an exhaustive database search with a registered protocol nor an original systematic review, and it makes no claim to generate a new pooled effect estimate.
The evidentiary synthesis is organized by the clinical and statistical question each study can actually answer. Descriptive surveys estimate experiences under a particular definition. Prospective cohorts describe timing and associations. Clinical trials identify the effect of assigning a retention intervention if follow-up and implementation permit. Qualitative interviews and case syntheses illuminate meanings and interactional sequences that administrative records obscure. Meta-analyses integrate published contrasts while inheriting the definitions and biases of the original studies. Official guidelines supply clinically relevant standards of follow-up and informed choice, but they are not themselves proof that every recommended practice improves retention.
Evidence is marked conceptually as descriptive, predictive, associational, experimentally comparative, or interpretive. A prognostic model that distinguishes groups at greater risk does not identify an ethically acceptable action that will reduce dropout. A positive correlation between alliance and attendance does not independently establish the direction of influence. These distinctions govern the interpretation of every numerical finding in the chapters that follow.
Conceptual Vocabulary and Operational Definitions
Treatment access begins before a first session. Help-seeking describes recognition of a perceived problem and the decision to approach potential assistance; referral completion describes the transition from referral to actual contact; initiation ordinarily means attending at least one treatment encounter. Engagement encompasses behavioral participation, collaborative involvement, perceived relevance, and continuing interaction with a service. Attendance is a count or proportion of scheduled or offered contacts, whereas adherence can refer to participation in the intervention's agreed procedures. Retention is the continued presence of an individual in a defined treatment pathway over time.
Dropout is frequently operationalized as stopping before a predetermined number of sessions, failing to attend a final appointment, failing to meet a clinician-defined course endpoint, or terminating before the provider believes treatment should end. These definitions identify partially overlapping groups. Treatment discontinuation is a broader transition: ending a modality, clinician relationship, prescription, episode, or service contact. Administrative discharge is a recorded organizational action and may occur after the patient has already disengaged. Loss to follow-up means an outcome or contact cannot be ascertained; it does not automatically show that the individual discontinued treatment.
A sound clinical classification should separate collaboratively planned completion, early satisfactory ending, agreed transfer, patient-initiated termination, clinician-initiated termination, interruption caused by resource constraints, and ambiguous contact loss. The patient and therapist may disagree about whether an ending is premature. In the international study by Wells (2013), dropout explicitly meant stopping before the provider wished; that definition must travel with the study's estimates.
Historical Foundations of Attrition Research
Psychotherapy attrition was investigated long before the contemporary language of implementation science. Earlier research used different thresholds for missed visits or completion, sometimes treating an unsuccessful contact as a patient characteristic and sometimes as evidence of a mismatch between care and need. The 1993 meta-analysis by Wierzbicki (1993) brought together 125 studies and reported an average dropout estimate commonly cited as 47%. Its importance lies both in the magnitude of the concern and in making heterogeneity of definitions visible.
Barrett (2008) subsequently synthesized decades of early-withdrawal literature and emphasized that methodological ambiguity obstructed practice recommendations. The next major aggregate analysis, Swift (2012), gathered 669 studies and showed that overall estimates change with clinical setting, study design, and definitions. Its pooled estimate was markedly lower than the historical figure, but the difference cannot be assigned simply to a genuine secular improvement in retention.
A further historical development was the extension of attrition research from psychotherapy rooms to population surveys, routine-service records, and digital systems. The World Mental Health Surveys examined discontinuation across outpatient sectors and national income groups. Eysenbach (2005) argued that electronic health services require an explicit science of attrition. Recent prognostic research uses large clinical datasets, while qualitative synthesis renews attention to what patients and therapists experience during the breakdown of a treatment relationship. The field's history is thus a history of increasingly explicit measurement as well as expanding observation.
The Care Pathway: Entry, Participation, and Exit
The pathway to psychotherapy can be represented as eligibility, recognition or referral, offered contact, accepted contact, attendance, repeated participation, negotiated ending, and post-treatment follow-up. Every transition has a distinct denominator and mechanism. People who never receive an appointment are absent from clinic-based dropout statistics. People who attend once enter analyses that may have excluded the greater difficulty of initial access. Programs that count only enrolled users can appear highly retentive while failing to engage a large share of those referred.
Research should consequently report at least four stages: eligible or referred persons, persons offered care, persons initiating treatment, and persons continuing at prespecified landmarks. A fifth stage records the kind and reason for ending. These quantities answer different questions. A treatment can improve appointment attendance while leaving referral access unchanged; a service can broaden entry and thereby include people who face greater barriers to sustained participation. Comparing raw post-entry dropout rates without referral composition may punish the more inclusive service.
Time matters because hazards are seldom constant. The earliest sessions may have especially high rates of exit. A pre-specified follow-up period and observation of switching or re-entry help distinguish transient interruption from definitive exit. Waiting time, clinician continuity, contact mode, and insurance authorization affect exposure to each stage. Studying the pathway as a multistate process makes these features visible and encourages interventions matched to the particular transition under investigation.
Initiation and Nonattendance Before Therapy
Failure to begin treatment can arise from anticipated stigma, uncertain perceived benefit, mistrust, practical barriers, unavailable appointments, financial exposure, transportation, disability-related inaccessibility, competing work obligations, or a mismatch between the offered intervention and the person's goals. These explanations vary by service and population; none should be presumed for an individual from an administrative absence alone. A no-show records the failure of an appointment to occur, not the presence of a particular psychological disposition.
The distinction between initial uptake and later retention is fundamental for evaluating interventions. Providing better information might clarify options and reduce decisional conflict; flexible scheduling might reduce logistical failure; rapid contact after referral might preserve an opportunity to engage. Their effects must be estimated separately, ideally with recorded reasons for declined offers and attempts to follow up. Conditional comparisons only among those who initiate treatment can produce selection bias if the intervention itself changes who enters care.
Contemporary guidelines underscore the importance of involving the patient at treatment choice. NICE NG222 (2022; reviewed 2026) recommends matching treatment to clinical needs and preferences, discussing barriers, and making follow-up contact under specified circumstances. These are relevant clinical principles; they should be translated into testable service procedures rather than treated as a single universal explanation of initiation.
Engagement as a Multidimensional Process
Engagement includes observable contact, behavior during encounters, agreement about the purpose of care, use of strategies between sessions, and an evolving judgment that the work merits continued participation. A person can regularly attend while contributing little to collaborative decisions; another can miss a session for practical reasons yet remain deeply invested in recovery. Attendance, alliance, homework completion, treatment satisfaction, and symptom improvement should therefore be measured as related but distinguishable variables.
Behavioral participation is partly produced by the organization of care. The intervention's complexity, expected session frequency, cultural and linguistic accessibility, travel requirements, format, and opportunities for patient choice all influence the burden of participation. Treatment credibility and expectations influence whether participation seems worthwhile. The evolving therapeutic relationship affects whether difficulties can be raised before they become reasons to leave. External events, including housing and employment disruptions, can interrupt a highly valued therapeutic process.
For measurement, engagement can be specified at session, episode, and service levels, with different time windows. Session-level indicators include attendance, collaborative goal activity, and reported helpfulness. Episode-level indicators capture consistency, rescheduling, agreed modifications, and the final ending type. Service-level measures report uptake and retention across referral cohorts. Using a single undifferentiated engagement score sacrifices the information needed to identify which mechanism an intervention changed.
Premature Termination and Legitimate Early Ending
Premature termination is an interpretation of an ending relative to an expected therapeutic opportunity. Its evaluation depends on whose expectation is used, how the course of care was defined, whether sufficient goals had been reached, and whether a suitable alternative was arranged. A person whose difficulties improve rapidly may reasonably stop earlier than the protocol's designated endpoint. Another may discontinue because the intervention is ineffective, unsafe, alienating, or too burdensome. Automatically coding both as equivalent treatment failure suppresses clinically relevant information.
At the same time, treating every patient-defined ending as successful would obscure substantial unmet need. Ending quality should be examined through patient-reported goal attainment, symptom and functioning change, expressed reasons, adverse experiences, and subsequent care when obtainable. These measures need a procedure that does not make a positive assessment conditional on continued cooperation with the original service.
Ethically competent research respects the right to decline and leave treatment while investigating whether services are imposing avoidable costs or failing to respond to distress. The core distinction is between the factual event of discontinuation and the evaluative conclusion that the ending occurred too early for the person's needs. Both variables belong in a mature evidence system.
What Meta-Analytic Dropout Rates Measure
Swift (2012) estimated a weighted dropout rate of 19.7% with a 95% confidence interval from 18.7% to 20.7% across 669 studies involving 83,834 clients. These figures are important as a historical cross-study synthesis, not as the expected rate for every clinic. The authors found moderation by diagnosis, age, therapist experience, service setting, operational definition, and efficacy versus effectiveness design. Such variation is intrinsic to the object of study, because the event itself is partly defined by treatment expectations.
The much-discussed 47% from the 1993 review cannot be subtracted from 19.7% to produce an unbiased estimate of improvement over two decades. Different study samples, termination rules, programs, clinical populations, and analytical methods contribute to the contrast. Similarly, studies of a single disorder or modality may yield estimates far from either overall figure. Cooper (2015) focused their meta-analysis on individual psychotherapy for unipolar depression precisely because condition-specific evidence requires a narrower denominator.
Every dropout percentage should be accompanied by its treatment, setting, population, endpoint definition, follow-up duration, number at risk, and missing-data assumptions. Pooled estimates deserve reported heterogeneity and sensitivity analyses. A high rate can reflect substantial exclusion or unsuitable care, while a low rate can reflect short treatment duration, selective enrollment, or a weak threshold for counting completion. Clinical interpretation follows the design rather than the headline percentage.
International and Service-Level Variation
The World Mental Health Survey analysis by Wells (2013) used representative household surveys in 24 countries. Among 8,482 respondents reporting outpatient mental health treatment in the previous year, the study defined dropout as stopping before a provider wanted and reported 31.7% overall. The estimates differed by country income category and by treatment sector, and earlier visits carried heightened risk of discontinuation. These data demonstrate that the phenomenon extends beyond formal psychotherapy trials into the organization of routine mental health care.
Cross-national comparisons require care because treatment pathways, provider supply, insurance systems, cultural meanings of care, recall patterns, and thresholds for provider-desired continuation differ. The household survey design reaches people outside participating clinics but depends on retrospective self-report. Clinic records provide session precision but often cannot identify whether a patient subsequently obtained effective care elsewhere. Each source compensates for a weakness of the other without eliminating it.
The most useful comparative unit is therefore a well-described service cohort embedded within a particular care system. A scientifically justified international program would link patient-reported reasons, appointment data, regional service availability, and clinically meaningful outcome measures, then examine whether subgroup differences persist after standardizing the definitions and opportunities for contact.
The First Session and the Hazard of Early Exit
The beginning of therapy demands a rapid alignment of expectations, perceived safety, professional credibility, available resources, and workable aims. A person may attend the first session to learn whether treatment fits at all. For research, that early deliberation is part of the observed process, not an error to be excluded. It can also be a period of elevated clinical vulnerability when presenting problems are severe and service continuity matters greatly.
In a large community-setting prognostic study, Giber (2026) examined electronic records from 40,732 adults with depression receiving at least one individual psychotherapy session in Massachusetts care networks. Their paper reported 28.4% dropout under the study's definition and showed that a large share of observed exits occurred after the initial visit. The study adds scale and contemporary routine-care relevance but addresses a specific population and record-based definition, rather than all psychotherapies everywhere.
A scientifically defensible first-session approach combines clear explanation of the proposed work, invitation to discuss alternatives, assessment of immediate barriers, mutually understood next steps, and attention to possible safety needs. Which component reduces dropout, for whom, and at what cost remains an empirical intervention question. Early contact should support an informed decision about therapy, including an informed decision to decline a particular service.
Therapeutic Alliance and Collaboration
The therapeutic alliance conventionally includes agreement on goals, agreement on tasks, and an interpersonal bond. It offers a plausible framework for understanding why patients persist through difficult phases of treatment, but its components require specific measurement. Sharf (2010) meta-analyzed 11 studies of adult individual psychotherapy, involving 1,301 participants, and found a moderate association between weaker alliance and dropout (reported d = 0.55). The evidence identifies a meaningful relationship, subject to the usual difficulty of confounding and reverse causation.
Alliance can deteriorate because therapy is ineffective, because a patient is considering leaving, or because a mismatch in goals has become explicit. Symptoms and external constraints can influence both alliance ratings and attendance. More broadly, the large Flückiger (2018) synthesis established the relevance of alliance to psychotherapy outcomes, while the subsequent Flückiger (2020) addressed whether alliance-outcome relationships survive accounting for measured intake and concurrent processes. These findings concern outcome, not solely dropout, and should be applied to retention through a distinct evidentiary argument.
The clinical implication is to measure collaboration early and repeatedly, invite disagreement, and examine whether the proposed work still fits the person's understanding of their difficulties. An alliance score alone is insufficient. Particularly when relationships deteriorate, research should examine the conversation and responses that follow rather than assume that every low score calls for the same scripted intervention.
Therapeutic Misattunement and Rupture
Dropout can develop through small interactional events that accumulate into an unworkable treatment relationship. A clinician might proceed with a technique before establishing its rationale, misunderstand a patient's cultural reference, interpret reluctance as resistance, or fail to notice disagreement about goals. These events are not reliably recoverable from a final session count. Qualitative methods are valuable because they reconstruct who experienced a difficulty, when it occurred, and how participants interpreted the subsequent response.
Notaerts (2025) synthesized 11 published psychotherapy single-case studies describing dropout. Their seven reported themes included misattunement, inert relationships, unmanaged therapist responses, unmanaged interventions, rigid protocol adherence, readiness for change, and recurring interpersonal dynamics. The synthesis is rich in process detail but selects studies that published substantive case accounts of dropout. It cannot establish population prevalence or the probability that any named theme will cause discontinuation in routine practice.
The complementary clinician survey by Kullgard (2022) asked 116 Swedish practitioners about dropout and perceived reasons, including dissatisfaction and unmet expectations. Its convenience recruitment and clinician perspective are explicit limits. Together these sources show why patient, therapist, and independent perspectives should be collected wherever feasible. Rupture is a possible process explanation, but the scientific test requires temporal observation and comparison with similar encounters that continue.
Expectancy, Credibility, and Perceived Benefit
A treatment is sustained partly through an evolving prediction that further participation is worth the investment. Patients interpret initial explanations, early experiences, feedback on progress, and changes in everyday functioning to decide whether continuing makes sense. Expectations may concern relief, understanding, a therapeutic relationship, social consequences, or skill acquisition. When an intervention has a delayed effect, the absence of immediate change can be interpreted as nonresponse unless the nature and time course of the treatment have been adequately discussed.
Expectation and credibility should be considered dynamic rather than fixed attributes. Symptoms may worsen temporarily, a particular task may feel alien, or a life event may change what the person can afford to do. A clinician's explanation may improve understanding, but optimistic framing without recognition of uncertainty can erode trust if the expected response fails to occur. The science of engagement should ask whether repeated credible assessment and collaborative adjustment predict continued attendance beyond baseline preferences and severity.
A principled design measures pretreatment expectancy, reassesses it after early sessions, records perceived usefulness and adverse experiences, and observes subsequent attendance. Causal claims require testing interventions that modify information or shared planning while accounting for changes in actual treatment content. Retention should not be pursued through promises that exaggerate likely benefit.
Symptoms, Diagnoses, and Clinical Complexity
Clinical presentation can alter the practical and psychological demands of entering treatment. Depression can affect energy and initiative; anxiety can make travel, social interaction, or anticipated exposure aversive; substance-related problems can disrupt routines; enduring interpersonal difficulties can complicate collaboration. These are hypotheses about pathways, not deterministic predictions based on labels. Clinical diagnosis explains only part of a person's opportunity and motivation to attend.
Swift (2012) reported diagnosis-related moderation in the aggregate literature. De Salve (2025) reviewed predictors of psychotherapy dropout in personality disorders across 22 selected studies, reporting rates that ranged from 10.4% to 58% and identifying factors at patient, relationship, and service levels. Their findings again underscore heterogeneous definitions, modalities, and populations. Predictor associations should guide questions and research priorities rather than justify denying treatment to people regarded as difficult to retain.
Complexity is best approached by considering interacting dimensions: presenting distress, functioning, exposure to adversity, prior experiences of care, feasibility of the treatment plan, and availability of coordinated support. Research designs should record co-occurring problems, treatment modifications, and adverse effects. They should also estimate whether accommodation of practical burdens benefits groups facing the greatest difficulties rather than assuming one standard schedule is feasible for everyone.
Socioeconomic Conditions, Access, and Equity
Retention is shaped by resources outside the consulting room. Transportation, paid leave, childcare, housing stability, income, digital access, disability accommodations, language, and geographic distribution of clinicians can determine whether a patient can act on a desire to participate. A person attending fewer sessions because they lose transportation and someone withdrawing because a treatment feels unhelpful contribute the same count to a crude dropout indicator but call for different system responses.
Observed group differences must be interpreted against the architecture of available care. The 2026 routine-care study by Giber (2026) found associations between recorded demographic or health-history variables and dropout; predictive associations do not license biological or psychological explanations of those demographic categories. Records may encode previous unequal opportunity, differences in referral pathways, and documentation bias. Using a model to prioritize support needs fairness evaluation and protections against exclusion.
Equity-oriented measurement should stratify referral opportunity, initiation, missed visits, and discontinuation where ethically and statistically appropriate. It should combine quantitative indicators with patient-generated accounts of service barriers and incorporate the cost of participation as an outcome. A program that raises average retention but increases the burden on people with the fewest resources can undermine its own public-health objective. An inclusive system treats access and continuity as shared properties of services and patients.
Development, Age, and Family Context
Engagement develops differently across childhood, adolescence, adulthood, and later life because decision authority, dependence on caregivers, school or work constraints, and service access change. For children, a parent or guardian may control scheduling and transport, while the young person has a different view of the treatment's relevance. Adolescents may face tensions among autonomy, confidentiality, family involvement, and institutional expectations. Older adults may encounter mobility, sensory, financial, or medical complexity affecting access to therapy.
Broad meta-analytic age associations cannot substitute for research specific to developmental context. The age moderator reported by Swift (2012) is a cross-study finding that also reflects the conditions and definitions present in the contributing literature. Within a younger clinical population, disengagement may be a family, school, clinic, and patient interaction rather than a single person's decision.
Developmentally informed studies should identify whose preferences were measured, who had practical control over attendance, whether confidentiality was understood, and whether the intervention was acceptable at a given stage. Longitudinal work can distinguish temporary interruption during transitions from more enduring loss of access. Decisions about care for minors require applicable safeguards and clinical oversight, and the research should report these contextual features transparently.
Therapist Practice and Organizational Conditions
Therapist experience, workload, communication, reliability, and responsiveness can contribute to differences in the course of care. A clinician's schedule and a clinic's cancellation policy determine whether missed appointments can be rescheduled; staffing continuity influences whether patients must repeatedly form new relationships. Such organizational variables may be correlated with patient case mix, making naive comparisons of individual therapists misleading.
Barrett (2008) highlighted the need to address early withdrawal as a practice problem that spans clinician and client characteristics. Swift (2012) observed variation associated with provider experience and setting. Neither source implies a fixed dropout disposition in particular clinicians. Research should separate patient-level variation, therapist-level clustering, and clinic-level constraints with multilevel models and clearly specified service characteristics.
Useful process measures include time to first appointment, cancellations attributable to clinicians, availability of alternative times, continuity across staffing transitions, and documented response to patient concerns. Interventions that train clinicians in collaborative expectations or improve service logistics can be compared experimentally, but fidelity and patient-centered outcomes must be assessed along with retention.
Digital Psychotherapy and the Law of Attrition
Digital interventions introduce new units of engagement: account creation, module initiation, activity completion, synchronous contact, message response, and sustained platform use. A user who stops opening an app is not necessarily equivalent to a patient who exits a therapeutic relationship. Some digital programs are intended for short, intermittent, or self-directed use. The denominator must include the prescribed or intended pattern of use, and meaningful benefit can sometimes occur with less engagement than designers anticipate.
Eysenbach (2005) called for explicit models of nonuse and study dropout in eHealth research. These are distinct outcomes: intervention nonuse is a behavioral feature of exposure, whereas loss of trial outcome data creates an inference problem. Karyotaki (2015) analyzed individual participant data from 2,705 people in ten randomized trials of self-guided web interventions for depression. The authors identified several predictors of treatment dropout in that specific program family, illustrating that self-guided engagement is structured by user and intervention characteristics rather than random variation alone.
Digital analytics can record detailed time stamps but often have weak information about meaning. A session opened for two minutes can reflect distraction, early improvement, or software failure. Measurement should therefore pair event logs with patient-reported goals, access conditions, symptoms, and reasons for stopping; a raw login count should not be treated as a universal dose of psychological treatment.
Human Support and the Supportive Accountability Model
Digital interventions sometimes incorporate guides, coaches, or clinicians who help participants formulate expectations, interpret activities, and overcome obstacles. Mohr (2011) proposed the supportive-accountability model, describing how accountability to a perceived trustworthy, benevolent, and competent coach might improve adherence. It is a theoretical model informed by adjacent literatures and clinical observations, rather than a direct demonstration that every added reminder or human contact causes better outcomes.
The model suggests testable mechanisms: clear and achievable expectations, reciprocal communication, credible support, and responsiveness to the patient's goals. A useful trial would distinguish the effects of contact frequency, coaching content, expertise, and accountability framing. It would also measure whether additional contact is welcomed or experienced as pressure. If reminders raise logins without increasing helpful engagement, the mechanism and clinical value remain uncertain.
Artificially mediated support adds further questions about perceived social presence, persistence of personalized guidance, accuracy of advice, transparency of automation, and escalation to human clinicians when risks arise. These questions belong to the empirical evaluation of intervention design; they do not establish that an artificial system has a human therapeutic relationship or clinical qualifications. Safety, consent, and human responsibility remain central when the platform addresses mental health difficulties.
Psychotropic Treatment and the Limits of This Dossier
Treatment discontinuation also appears in psychiatric medication research, but the constructs require careful separation. Stopping a psychological therapy, not taking prescribed medication, tapering medication with a clinician, and losing access to prescribing care have different causal pathways and potential harms. Medication adherence may be affected by tolerability, perceived effectiveness, cost, regimen complexity, and patient values. Pharmacological withdrawal and relapse are distinct medical questions with drug-specific evidence and regulatory requirements.
This monograph's central object is engagement with psychotherapy and outpatient psychosocial treatment. Medication is discussed where it influences continuity of integrated care or where parallel treatment makes the interpretation of psychotherapy dropout uncertain. Detailed pharmacokinetics, dose selection, withdrawal syndromes, and drug-specific stopping protocols belong to dedicated psychopharmacology and medication-adherence dossiers.
NICE NG222 distinguishes treatment review, monitoring, treatment concordance, and the safe management of stopping antidepressants. The guideline provides a concrete illustration of why an administrative record that a patient stopped care cannot establish whether that event was clinically planned or safe. Research data should document the modality being discontinued, ongoing prescribing relationships, and transfer or follow-up where relevant.
Attrition Measurement and Denominator Integrity
A reproducible dropout measure begins by defining an eligible cohort and the exact point at which risk starts. Studies should distinguish referral attrition from treatment attrition, count first-session nonreturn accurately, pre-specify when a treatment is expected to be complete, and classify transfer or patient-defined completion separately. Fixed session-count rules can be useful for protocol comparison but may distort person-centered therapy whose length legitimately varies.
Administrative definitions are attractive because they are standardized within a service. Yet the same missed contact can arise from clinician cancellation, patient choice, lost insurance coverage, hospitalization, or migration. A clinically oriented measure should supplement attendance with planned course length, patient and clinician accounts of ending, goal achievement, and documented reasons. Unknown reasons must be retained as unknown rather than silently assigned to a psychologically unfavorable category.
Investigators should report the denominator at each step, the time horizon, allowed rescheduling window, consequences of transfer, and whether people who never began care are excluded. Multi-episode patients pose a further problem: an individual may disengage from one therapist and re-enter treatment elsewhere. Properly distinguishing episodes from persons prevents a single person with several attempts from being counted as several independent patients in pooled analyses.
Time-to-Event Analysis and Competing Risks
Retention is intrinsically longitudinal. Counting termination by a fixed end date discards the timing of risk and produces different estimates for courses of different duration. Survival analysis models the time until a defined event while acknowledging incomplete follow-up. Kaplan–Meier estimates describe event-free survival under assumptions about censoring. Cox models estimate covariate associations with a hazard under specified proportionality assumptions. Both require a clearly defined time origin and event.
Not every ending competes on the same terms. Planned completion, clinician-initiated transfer, relocation, and unplanned patient departure can prevent observation of the particular dropout event of interest. Treating these as ordinary noninformative censoring may misrepresent cumulative incidence; competing-risk approaches can model their distinct occurrence. Repeated interruptions and returns may be better represented by multistate models, with transitions among active treatment, temporary pause, transfer, and exit.
Time-varying factors complicate analysis. Symptoms, alliance, scheduling feasibility, and treatment content change after therapy starts and may influence each other. Conditioning on a later variable affected by earlier treatment can create bias. A clear causal diagram and estimand should precede adjustment. Predicting departure within four sessions and estimating the effect of a service policy on departure are different questions that require different analytic strategies.
Dropout, Missing Data, and Treatment Effects
Dropout from therapy and dropout from research assessment are two different processes. A patient may stop therapy but complete follow-up interviews; another may finish every session yet fail to return outcome questionnaires. Confusing the two can bias estimates of intervention benefit. An intention-to-treat analysis preserves the randomized assignment contrast but still requires defensible handling of unobserved outcomes. Complete-case analyses can be especially misleading when symptom severity and care experience influence follow-up.
Missingness is conventionally evaluated under assumptions ranging from missing completely at random to missing at random conditional on observed variables and missing not at random. The last possibility is particularly plausible in psychotherapy research, where distress, dissatisfaction, and adverse events can affect willingness to provide further data. Multiple imputation or likelihood-based methods depend on models and observed information; neither automatically repairs informative missingness.
Sensitivity analyses should examine plausible departures from identifying assumptions, alternative definitions of termination, early versus late exits, and the impact of clinically important outcomes among those lost. Researchers should report how much evidence comes from individuals with complete data and whether dropout itself could have been influenced by assignment. The causal meaning of treatment efficacy depends on transparent follow-up, not only on a nominally sophisticated statistical method.
Risk Prediction, Machine Learning, and Clinical Utility
Predictive systems seek to identify people whose observed profiles resemble those of earlier patients who discontinued care. They can use symptoms, service utilization, appointment history, and other electronic records. Prediction can be operationally helpful when a reliable warning leads to a supportive, voluntary intervention. It does not by itself explain why a patient leaves or prove that the intervention changes the outcome.
Giber (2026) used logistic regression and random forests on a large Massachusetts electronic health record cohort to study dropout in depression care. The investigators observed associations between previous health-system contact and later discontinuation. Such models may encode institutional recording practices and historical inequity. Their usefulness depends on external validation across sites, temporal validation as care changes, calibration, error rates for relevant groups, and the actions attached to different risk estimates.
The ethically relevant target is equitable care improvement, not ranking patients by presumed cooperativeness. A model should never become a reason to withhold appointments from those predicted to leave. A credible evaluation would test whether offering an additional support option, triggered under a transparent policy, increases patient-valued continuity without causing stigma or administrative burden. The intervention trial is needed to move from accurate classification to demonstrable benefit.
Patient and Clinician Accounts of Why Care Ends
Reasons for dropout are not fully contained in attendance records. Patients may report a completed personal goal, lost trust, poor cultural fit, inconvenient scheduling, unmanageable cost, adverse effects, or a preference for different care. Clinicians may infer low motivation or premature improvement. Disagreement is itself informative: it reveals that the therapeutic enterprise did not have one shared account of what occurred.
The study of Kullgard (2022) surveyed clinicians and documented their perceived reasons for premature endings. The Notaerts (2025) qualitative meta-synthesis examined interactional process across published individual cases. Neither design supplies a representative estimate of patient-stated reasons in general services. Together they justify triangulation among perspectives rather than relying exclusively on the clinician's assessment or a coded administrative category.
High-quality exit research can invite brief, noncoercive patient feedback after a missed appointment, use independent interviewers where appropriate, and allow multiple reasons instead of forcing a single cause. It should protect confidentiality, avoid repeated unwanted contact, and provide meaningful options for future assistance. Qualitative categories can then guide quantitative hypotheses for prospective testing.
Role Induction and Early Treatment Preparation
Role induction is an early intervention that clarifies what the patient and clinician will do, how treatment is organized, what responsibilities are negotiable, and what challenges may arise. It can help patients interpret the first appointments and judge whether the proposed plan fits their needs. Its potential value has a stronger direct experimental basis than many generalized recommendations to improve engagement.
Swift (2023) meta-analyzed 17 studies of role induction in adult individual psychotherapy. Across the relevant subset, role induction was associated with a reduction in premature termination (15 studies; the authors reported an odds-ratio contrast of 1.64) and improved selected immediate or posttreatment outcomes. Their reported heterogeneity means an intervention's implementation and comparator remain relevant. An odds ratio must be interpreted in the direction actually coded by the study rather than converted casually into a guaranteed individual benefit.
A clinically appropriate role-induction procedure explains the planned mode and rationale, establishes patient rights, solicits preferences, discusses the possibility of nonresponse, and makes cancellations or changes of plan comprehensible. Researchers should test what content is essential and whether role induction helps people whose access barriers are largely structural. Clarification should be an aid to informed participation, not an instrument for extracting compliance.
Service Responses: Flexible Access and Continuity
An intervention portfolio should match the pathway where attrition occurs. Referral-stage barriers invite easier scheduling and clearer referral communication. First-session exits invite collaborative explanation, preference-sensitive selection, and early feedback. Midcourse interruptions may require changes to practical logistics, treatment intensity, therapist continuity, or the therapeutic plan. End-of-course discontinuation calls for an explicit assessment of whether care has achieved an agreed endpoint.
Service adaptations should be tested with appropriate controls and equity measures. Flexible scheduling can fail if appointment supply is unchanged. Digital access can help some patients while excluding others who lack reliable devices, privacy, connectivity, or digital literacy. Additional outreach can be supportive when chosen by the patient and intrusive when delivered without consent. The absence of a response to a reminder cannot establish unwillingness to receive care.
NICE NG222 recommends review and monitoring during depression treatment and provides guidance for people who miss follow-up in relevant contexts. Translating such principles into a particular service requires local safety procedures and measurable outcomes. A successful program should improve appropriate continuity, patient understanding, and functioning rather than maximizing the number of appointments at any cost.
Safety, Deterioration, and Appropriate Exit
An unplanned interruption can matter especially when someone has acute risk, rapid symptom deterioration, inability to meet basic needs, or a treatment-related adverse experience. Clinical responsibility depends on the setting, clinician role, applicable law, and individual assessment. Research on retention should incorporate severe outcomes and harms without treating all patient departures as emergencies or assuming that additional attendance always protects patients.
Clinical systems may require agreed plans for follow-up, continuity and escalation according to evaluated risk. These plans should address missed contact, transfer across services, and communication with appropriate professionals while respecting privacy and consent. A research publication cannot substitute for individualized clinical assessment. In contexts involving medication, safe review of changes and withdrawal requires relevant qualified care and official guidance, as illustrated by NICE NG222.
The safety outcome must be separated from the retention endpoint. A patient may appropriately discontinue a harmful or unsuitable therapy and experience improved safety. Conversely, a patient who becomes unreachable while deteriorating may need a carefully proportionate service response. Future trials should pre-specify adverse event capture and independent review when appropriate, and should report how contact policies operate for individuals who prefer not to continue.
Ethics of Engagement and the Right to Withdraw
Engagement research is ethically meaningful only when participation remains voluntary wherever the clinical and legal context permits choice. The purpose of improving retention is to make effective care available and acceptable, with the person's interests and informed preferences at the center. A policy that increases session counts by making cancellation difficult or by pressuring patients can improve an administrative metric while degrading quality.
Patient preferences have empirical support for association with retention in Windle (2020) and the work of Swift (2018). Their findings are compatible with an ethical conception of treatment that values agency and collaborative choice. Preferences are sometimes constrained by clinical safety, available resources, or evidence of likely harm; competent shared decision-making acknowledges those constraints explicitly.
Patient-reported experience should therefore be a core endpoint alongside retention. Meaningful reporting may include decisional understanding, goal alignment, perceived respect, cost and effort, clinical change, and accessibility of alternative care. Research governance should protect the right to decline interviews about reasons for leaving, establish data-minimization practices, and prevent risk algorithms from becoming mechanisms of exclusion.
Clinical and Statistical Interpretation of Evidence
The most credible propositions in this field concern recurring associations and defined intervention contrasts. Large meta-analyses show substantial treatment discontinuation but also considerable measurement heterogeneity. Treatment preference is consistently associated with reduced dropout, and role induction has been studied as an intervention with favorable pooled findings. Weak alliance is associated with dropout; the direction and mechanism of influence require stronger longitudinal and experimental tests. Digital nonuse is a pervasive measurement and implementation challenge, not a uniform diagnosis of failed treatment.
An evidence table should distinguish four columns: question, design, observed estimate, and principal inference limit. For example, Swift (2012) provide a broad pooled descriptive estimate, whereas Windle (2020) compare preference conditions. Wells (2013) provide cross-national survey coverage. Giber (2026) offer high-dimensional prediction within documented health systems. Notaerts (2025) provide qualitative process interpretations.
A coherent synthesis does not assign one global certainty level to all these results. It asks which inference is possible from each design, what population is covered, and whether the phenomenon was measured in a way that corresponds to the scientific question. This discipline is especially important when policy decisions have consequences for people who already experience barriers to mental health care.
Artificial Era and New Forms of Treatment Engagement
In the Artificial Era, human treatment pathways increasingly intersect with automated scheduling, digital symptom monitoring, conversational interfaces, and algorithmic decision support. Their effect on retention can be studied at the level of observable service operations: did an offered reminder reach the patient, did tailored information improve understanding, did a transparent automated triage tool facilitate appropriate human contact, and did system design alter which groups could access care? These questions support empirical testing without requiring speculation about the internal experience of the artificial component.
Artificial agency in service infrastructure can be operationalized as the ability of a system to execute bounded decisions under specified policies, with audit logs, error handling, and human oversight. Relational continuity can be measured through consistent communication, recall accuracy within consented boundaries, and reliable handoff among clinical actors. A digital system may contribute to psychological care through organization and communication, while responsibility for clinical treatment decisions remains appropriately assigned to qualified persons and institutions.
A productive research program would randomize defined supportive functions and evaluate uptake, attendance, patient experience, clinical outcomes, safety incidents, accessibility, and privacy costs. The emerging field should avoid assuming that more digital interaction is universally desirable. An artificial system's success in this domain is measured by whether it improves informed, equitable, and safe care continuity.
Research Priorities and Unresolved Problems
The first unresolved problem is definitional. Studies need harmonized categories that preserve the differences among never initiating treatment, interrupted attendance, collaboratively agreed completion, switching, unexpected termination, and unobservable outcome status. A second problem concerns denominators: estimates restricted to first-session attendees cannot measure the total access burden faced by eligible populations. A third is temporality: alliance, symptoms, expectations, and practical barriers change together and call for frequent, well-timed measurement.
Fourth, retention interventions need direct randomized or credible quasi-experimental tests in routine settings. The field needs trials that compare specific components of role induction, preference accommodation, practical assistance, and personalized follow-up, with prespecified clinical benefit, consent, and equity outcomes. Fifth, predictive modeling requires external validation and intervention studies. Risk scores have little service value without evidence that their use improves care for those identified.
Sixth, research should expand patient-generated reasons for leaving, using multilingual and culturally responsive methods, and link them to service records without compromising confidentiality. Seventh, digital research needs a principled definition of meaningful use in programs designed for intermittent engagement. Eighth, adverse effects and appropriate patient-led termination should be measured as primary distinctions rather than erased by a retention target. Finally, a cumulative science of discontinuation requires transparent reporting, reusable operational definitions, open analytic code where feasible, and replication across health systems and populations.
Conclusion
Engagement, dropout, and treatment discontinuation form a connected scientific field concerned with how psychological care becomes available, meaningful, sustainable, and appropriately concluded. The literature establishes that discontinuation is common but definition-sensitive, that risk is distributed across patients, relationships, clinicians, and service systems, and that early treatment stages deserve particular attention. Evidence for accommodating preferences and preparing patients for therapy offers actionable directions, while observational predictors and alliance associations require appropriately bounded interpretation.
The central synthesis is that the unit of explanation should be the evolving patient–treatment–service relationship. A missed session is one observation within this sequence, not a complete account of its cause or clinical meaning. Research progress depends on following the entire care pathway, measuring reasons and outcomes, preserving legitimate exits, and evaluating interventions by patient-valued benefit as well as attendance. In this framework, improved continuity means that people can enter fitting care, influence its course, and leave or transfer with informed support and appropriate safety protections.
References
- Barrett, M. S., Chua, W.-J., Crits-Christoph, P., Gibbons, M. B., Casiano, D., & Thompson, D. (2008). Early withdrawal from mental health treatment: Implications for psychotherapy practice. Psychotherapy, 45(2), 247–267. https://doi.org/10.1037/0033-3204.45.2.247
- Cooper, A. A., & Conklin, L. R. (2015). Dropout from individual psychotherapy for major depression: A meta-analysis of randomized clinical trials. Clinical Psychology Review, 40, 57–65. https://doi.org/10.1016/j.cpr.2015.05.001
- De Salve, F., Rossi, C., Gioacchini, E., Messina, I., & Oasi, O. (2025). Dropout in psychotherapy for personality disorders: A systematic review of predictors. Clinical Psychology & Psychotherapy, 32(3), e70080. https://doi.org/10.1002/cpp.70080
- Eysenbach, G. (2005). The law of attrition. Journal of Medical Internet Research, 7(1), e11. https://doi.org/10.2196/jmir.7.1.e11
- Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The alliance in adult psychotherapy: A meta-analytic synthesis. Psychotherapy, 55(4), 316–340. https://doi.org/10.1037/pst0000172
- Flückiger, C., Del Re, A. C., Wlodasch, D., Horvath, A. O., Solomonov, N., & Wampold, B. E. (2020). Assessing the alliance-outcome association adjusted for patient characteristics and treatment processes: A meta-analytic summary of direct comparisons. Journal of Counseling Psychology, 67(6), 706–711. https://doi.org/10.1037/cou0000424
- Giber, B., Sharma, A., Broadie, D., McCoy, T. H., Jr., Gunning, F. M., Perlis, R. H., Doshi-Velez, F., & Solomonov, N. (2026). Identifying patients at risk for psychotherapy dropout in community settings. JAMA Network Open, 9(8), e2631565. https://doi.org/10.1001/jamanetworkopen.2026.31565
- Karyotaki, E., Kleiboer, A., Smit, F., Turner, D. T., Pastor, A. M., Andersson, G., Berger, T., Botella, C., Breton, J. M., Carlbring, P., Christensen, H., de Graaf, E., Griffiths, K., Donker, T., Farrer, L., Huibers, M. J. H., Lenndin, J., Mackinnon, A., Meyer, B.,. Cuijpers, P. (2015). Predictors of treatment dropout in self-guided web-based interventions for depression: An individual patient data meta-analysis. Psychological Medicine, 45(13), 2717–2726. https://doi.org/10.1017/S0033291715000665
- Kullgard, N., Holmqvist, R., & Andersson, G. (2022). Premature dropout from psychotherapy: Prevalence, perceived reasons and consequences as rated by clinicians. Clinical Psychology in Europe, 4(2), e6695. https://doi.org/10.32872/cpe.6695
- Mohr, D. C., Cuijpers, P., & Lehman, K. (2011). Supportive accountability: A model for providing human support to enhance adherence to eHealth interventions. Journal of Medical Internet Research, 13(1), e30. https://doi.org/10.2196/jmir.1602
- National Institute for Health and Care Excellence. (2022). Depression in adults: Treatment and management (NICE Guideline NG222; reviewed January 30, 2026). https://www.nice.org.uk/guidance/ng222
- Notaerts, L., De Smet, M., Finn, M., Van Nieuwenhove, K., Hennissen, V., & Meganck, R. (2025). It takes two to tango: A qualitative meta-synthesis on processes of psychotherapy dropout from the Single Case Archive. Research in Psychotherapy: Psychopathology, Process and Outcome, 28(3), 876. https://doi.org/10.4081/ripppo.2025.876
- Sharf, J., Primavera, L. H., & Diener, M. J. (2010). Dropout and therapeutic alliance: A meta-analysis of adult individual psychotherapy. Psychotherapy, 47(4), 637–645. https://doi.org/10.1037/a0021175
- Swift, J. K., & Callahan, J. L. (2009). The impact of client treatment preferences on outcome: A meta-analysis. Journal of Clinical Psychology, 65(4), 368–381. https://doi.org/10.1002/jclp.20553
- Swift, J. K., Callahan, J. L., Cooper, M., & Parkin, S. R. (2018). The impact of accommodating client preference in psychotherapy: A meta-analysis. Journal of Clinical Psychology, 74(11), 1924–1937. https://doi.org/10.1002/jclp.22680
- Swift, J. K., & Greenberg, R. P. (2012). Premature discontinuation in adult psychotherapy: A meta-analysis. Journal of Consulting and Clinical Psychology, 80(4), 547–559. https://doi.org/10.1037/a0028226
- Swift, J. K., Penix, E. A., & Li, A. (2023). A meta-analysis of the effects of role induction in psychotherapy. Psychotherapy, 60(3), 342–354. https://doi.org/10.1037/pst0000475
- Wells, J. E., Browne, M. O., Aguilar-Gaxiola, S., Al-Hamzawi, A., Alonso, J., Angermeyer, M. C., Bouzan, C., Bruffaerts, R., Bunting, B., Caldas-de-Almeida, J. M., de Girolamo, G., de Graaf, R., Florescu, S., Fukao, A., Gureje, O., Hinkov, H. R., Hu, C., Hwang, I., Karam, E. G.,. Kessler, R. C. (2013). Drop out from out-patient mental healthcare in the World Health Organization's World Mental Health Survey initiative. British Journal of Psychiatry, 202(1), 42–49. https://doi.org/10.1192/bjp.bp.112.113134
- Wierzbicki, M., & Pekarik, G. (1993). A meta-analysis of psychotherapy dropout. Professional Psychology: Research and Practice, 24(2), 190–195. https://doi.org/10.1037/0735-7028.24.2.190
- Windle, E., Tee, H., Sabitova, A., Jovanovic, N., Priebe, S., & Carr, C. (2020). Association of patient treatment preference with dropout and clinical outcomes in adult psychosocial mental health interventions: A systematic review and meta-analysis. JAMA Psychiatry, 77(3), 294–302. https://doi.org/10.1001/jamapsychiatry.2019.3750