Implementing modular competence-oriented psychotherapy in inpatient psychiatry: a three-year analysis of restraint and seclusion

Abstract

Background

Coercive measures remain used in psychiatric care, despite ongoing ethical concerns and efforts to minimize their application. Psychotherapeutic or competence-oriented frameworks may contribute to reducing situations that lead to coercion, yet empirical evidence is scarce.

Objective

This naturalistic study examined whether the introduction of Modular Competence-Oriented Psychotherapy (MCP) at site A was associated with changes in the frequency and duration of coercive measures (2022–2024), compared to site B providing treatment as usual.

Methods

Routinely collected inpatient data from two psychiatric sites in Germany were analyzed retrospectively (n = 15,361 cases). Generalized linear models assessed annual changes in the use of restraint and seclusion. Linear models examined changes in the duration of restraint and seclusion episodes.

Results

Overall, the probability of coercive measures was significantly higher at site B than at site (A). At site A, the likelihood and duration of restraint and the duration of seclusion significantly declined over time, while the likelihood of seclusion remained stable. Except for seclusion duration, time trends did not differ significantly between sites. For seclusion duration, site A showed a significantly stronger decline than site (B). Descriptively, assaults decreased at site A but remained stable at site B.

Conclusions

Implementation of MCP at site A was associated with reductions in the duration of restraint and seclusion, without indications of increased risk events. Although the naturalistic design and concurrent changes at site B preclude causal conclusions, the findings offer preliminary evidence for the feasibility and potential value of integrating a competence-oriented psychotherapeutic framework in psychiatry.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-026-08468-1.

Introduction

The use and benefit of coercive measures in psychiatry, including forced medication, seclusion and physical restraint, remain the subject of considerable debate in both the scientific literature and clinical practice [1–3]. The controversy is reinforced by findings showing that the use of coercive measures is often determined by ideological decisions and varies significantly across countries in the European Union [4, 5]. On the one hand, coercive measures should be applied to protect individuals in emergency situations in which they pose a danger to themselves and/ or others [6]. On the other hand, such measures restrict the personal freedom of the individual concerned and conflict with core ethical principals in psychiatry, such as patient autonomy [1, 7]. Furthermore, evidence suggests that coercive measures are frequently applied without direct benefit to the individual concerned [6, 8]. Such measures may result in substantial adverse outcomes for both patients and clinical staff, manifesting in negative short- and long-term consequences (e.g. post-traumatic stress symptoms) [1]. In addition, the majority of patients perceive coercive measures as humiliating and punitive (for an overview see 1). Therefore, it is essential to reduce the use of coercive measures to an absolute minimum and apply them only as a last resort [6, 8, 9]. Against this background, the present study investigates whether the implementation of a psychotherapeutic framework within routine psychiatric care can contribute to reducing the use and duration of coercive measures.

One effective way to reduce the use of coercive measures is to train psychiatric staff to deescalate dangerous situations before coercion becomes unavoidable [7]. Therapeutic interventions such as validation and commitment strategies can be helpful in these situations, as they strengthen staff members’ sense of self-efficacy and convey to the patient that the staff understands and is able to help them. Two prominent examples of evidence-informed conflict reduction strategies are the Safewards Model [10, 11] and the Six Core Strategies [12, 13]. Evidence supporting these team-based interventions comes from several studies demonstrating that efforts to train psychiatric teams in negotiation and de-escalation skills are effective and worthwhile [14, 15].

Moreover, there is evidence that single interventions (e.g., team-reflection on previously implemented coercive measures) can reduce the number of coercive measures in psychiatry, especially when these interventions are individualized and tailored to the needs of the specific ward [15]. Such single (vs. complex) interventions can be more easily implemented in psychiatric care, as time constraints and patients with high care needs often limit the resources available for new initiatives. A comprehensive collection of both single and complex interventions has been provided by the German Society for Psychiatry, Psychotherapy, and Psychosomatics. They developed the clinical practice guidelines Prevention of Coercion: Prevention and Treatment of Aggressive Behavior in Adults [9], which comprise 89 recommendations and statements based on evidence and expert-consensus [9]. Initial evidence for the feasibility and effectiveness of these clinical practice guidelines has been provided by a randomized controlled trial [16].

Beyond these strategies specifically designed to prevent or reduce the use of coercive measures, it is also plausible to address the underlying behaviors that result in the use of coercion. Aggressive, violent, and uncooperative behavior can be understood as dysfunctional strategies used by patients to express their personal boundaries or to assert what they want or believe is their right in a given situation. Several effective therapies, particularly those belonging to the “third wave” of cognitive-behavioral approaches, address the reduction of dysfunctional behaviors in a greater therapeutic setting [17]. For example, Dialectical Behavior Therapy (DBT; [18]) is designed to treat individuals with personality dysfunction, especially those with borderline personality disorder. It has been shown to reduce maladaptive behaviors such as non-suicidal self-injury, suicidality, anger, and aggression [19–21]. Conceptualizing violent aggression as a dysfunctional behavior suggests that its treatment within a psychotherapeutic setting is both plausible and promising. However, a large proportion of psychiatric clinics still do not meet the standards for consistent psychotherapeutic care [22]. From both a clinical and economic perspective, it is therefore reasonable to implement psychotherapeutic approaches within psychiatric wards that explicitly address dysfunctional behaviors such as aggression.

A recently developed therapeutic approach that aims to promote the use of psychotherapy in psychiatry through systematic training in key therapeutic strategies is the Modular Competence-Oriented Psychotherapy (MCP; German: modulare kompetenzorientierte Psychotherapie (MKP); 22, 23). MCP is designed for multidisciplinary psychiatric teams treating patients across the full spectrum of psychiatric disorders, including those in acute crisis. It provides intervention strategies embedded in a modular structure for the whole team that can be applied continuously throughout the entire treatment period. Thus, from admission to discharge, the team follows a shared therapeutic framework, enabling every member of the team to use the same therapeutic interventions and reinforce the same adaptive behaviors in patients around the clock [22].

MCP integrates basic transdiagnostic and cross-professional treatment strategies that are shared across established psychotherapeutic approaches, particularly Cognitive Behavioral Therapy [22, 23]. Core elements include therapeutic communication techniques such as validation and commitment, principles for effective teamwork, and structured treatment processes (see Fig. 1). The transdiagnostic modules represent basic psychotherapeutic competencies (see also [24]) and enable teams to implement the five factors of psychotherapy effectiveness [25] within a competency-oriented and process-based framework [26, 27], allowing flexible adaptation of therapeutic strategies to patients’ comorbidities and fluctuating symptom profiles. An initial feasibility study of the pilot version of MCP showed high acceptance among psychiatric staff [28]. For more detailed information about the content of MCP, see the manual [23], and for more information about the development process, see Franz and Franz [22].

Fig. 1.

Overview of modules in the Modular Competence-Oriented Psychotherapy. Basic modules: cross-professional, transdiagnostic and independent of comorbidities. Competence-oriented modules: Applied across all clinical units without a specific psychotherapeutic focus, including crisis intervention. In specialty areas with disorder-specific therapies (e.g., depression, BPD, psychosis), used primarily for treating comorbidities - depending on individual competence and profession-specific scope

Although MCP was not originally designed to reduce coercion, it provides a transparent, patient-centered framework for personalized recovery that may help prevent aggression and related crises by fostering a respectful therapeutic attitude, consistent problem and functional analyses, and contingency management [23]. The latter links individually defined treatment goals to structured, predictable reinforcement of adaptive behaviors, thereby promoting predictability and autonomy for patients and reducing staff members’ perceived need for external control. This supports shared decision-making between patients and the therapeutic team. Consistent with empirical findings that greater patient engagement, perceived control, and behavioral self-regulation are associated with lower rates of coercive measures [1], MCP’s individualized, skill-focused, and contingency-informed structure may represent a promising framework to strengthen patient-centered care while reducing reliance on coercive interventions.

The present study

The present study assessed whether the implementation of MCP is associated with a reduction in the frequency and duration of coercive measures in psychiatric care in comparison to treatment as usual (TAU). Routine clinical care data assessing coercive measures at two sites of the Vitos Clinic Giessen-Marburg (Germany) were analyzed. At site A, MCP was implemented in the second half of 2022, whereas site B continued using TAU. It is important to note, however, that the introduction of MCP at site A and the first changes according to this became known to staff at site B during the observation period, due to shared organization structures. Therefore, unplanned behavioral adaptations at site B cannot be ruled out. Previous research indicates that such cross-site awareness can influence behavior through expectancy or organizational learning mechanisms [16, 29]. This contextual factor is therefore relevant for interpreting our retrospective comparisons between the two sites.

To compare the two sites (site A with MCP and site B with TAU) we formulated a set of hypotheses as presented below. As an initial step, to assess overall differences between site A and site B, we estimated the probabilities of experiencing physical restraint or seclusion at each site (aggregated over the period 2022–2024) and conducted a between-group comparison. Building on these preliminary analyses, we tested our specific hypotheses:

First, we hypothesized that the likelihood of physical restraint use significantly declines at site A from 2022 to 2024, as teams become more familiar with MCP over time (H1a). Moreover, we expected the decline of physical restraint use from 2022 to 2024 at site A to be significantly stronger compared to site B with TAU (H1b).

Second, we hypothesized that the duration (in hours) of restraint per case significantly declines over the years 2022–2024 at site A (H2a). In addition, we hypothesized that this decline from 2022 to 2024 is stronger at site A using MCP in comparison to site B using TAU (H2b).

Third, we expected the equivalent pattern for the use of seclusion. We hypothesized that the likelihood of being secluded during the treatment period significantly declines from 2022 to 2024 at site A, as team members become more familiar with MCP (H3a). This reduction from 2022 to 2024 should be significantly stronger at site A in comparison to site B (H3b).

Lastly, we hypothesized that the duration of seclusion per case (in hours) declines at site A from 2022 to 2024 (H4a). This decline should outperform a potential decline at site B (H4b) from 2022 to 2024.

Method

We retrospectively analyzed routine clinical care data collected from January 2022 to December 2024 at two sites of the Vitos Clinic Giessen-Marburg (Germany). We included all inpatient wards at the two sites (site A n = 9, site B n = 5). The included data were provided by the medical controlling department and were pseudonymized. These retrospective analyses are part of an evaluation study on the effectiveness and feasibility of MCP in collaboration with the University of Leipzig (ethics approval was granted by Landesaerztekammer Hessen; LAEKH: 2022-2943-evBO, in accordance with the Declaration of Helsinki).

The study sites exhibit highly comparable structural characteristics, including ward arrangement, park-like surroundings, peripheral location, staffing ratios (see Additional File 1) and average hospital days (see Table 2). Both sites operate under mandatory service provision with an obligation to admit all patients, precluding the selection of less severely ill individuals, and they serve similar catchment areas, including deprived areas / social hotspots. The clinical focus is also comparable across sites, encompassing psychotic disorders, substance use disorders, geriatric psychiatry, depression, borderline personality disorders (BPD), and trauma-related disorders. At site B, BPD and trauma are treated within a single ward, whereas at site A these specialties are managed in separate wards.

Table 2.

Descriptive values of average hospital days, assaults, restraint, seclusion and suicidality at the two sites A and B (year 2022–2024) in absolute values and percentages based on the total number of cases

	Site A (n = 10,041)						Site B (n = 5320)
	2022		2023		2024		2022		2023		2024
	n/ M	%/SD	n/ M	%/ SD	n/ M	%/SD	n/ M	%/SD	n/ M	%/ SD	n/ M	%/ SD
Average hospital days	21.04	(22.44)	20.82	(23.12)	22.27	(23.54)	21.32	(22.90)	24.01	(25.27)	25.03	(28.51)
Assaults	100.00	(2.99)	94.00	(2.64)	51.00	(1.63)	108.00	(5.89)	91.00	(5.16)	101.00	(5.87)
Restraint	102.00	(3.06)	108.00	(3.03)	68.00	(2.17)	125.00	(6.81)	122.00	(6.92)	91.00	(5.28)
Average restraint per case (hrs.)	42.45		31.31		24.02		45.90		40.37		35.66
Seclusion	126.00	(3.77)	149.00	(4.18)	130.00	(4.15)	116.00	(6.32)	126.00	(7.15)	124.00	(7.20)
Average seclusion per case (hrs.)	181.78		149.01		71.61		131.54		144.76		151.87
Suicide attempts	4.00	(0.12)	6.00	(0.17)	3.00	(0.09)	2.00	(0.11)	8.00	(0.45)	11.00	(0.63)
Suicide	1.00	(0.03)	2.00	(0.06)	0.00		1.00	(0.05)	0.00		1.00	(0.06)

MCP implementation procedure

The implementation of the MCP framework at site A started in the second half of 2022 with an onboarding workshop for all available staff members at each ward separately. During this workshop, an overview of the basic modules of MCP was taught by an advanced MCP therapist (see Fig. 1 for an overview of the basic modules). Afterwards, the wards themselves were responsible for the training of their staff and the implementation of the basic modules, but reported their progress in regular intervals during leadership team meetings. As most staff members in psychiatric settings work in 24-hour shifts, it was not possible to train all members of the team within the same sessions. Therefore, we decided to provide online learning materials (e.g., multiple videos of teaching sessions for every basic module, handouts) to all staff members to facilitate mobile working and learning. Additionally, the wards organized joint viewing of teaching sessions in weekly intervals and in-person workshops on the basic modules on a voluntary basis, to facilitate in-depth learning. Furthermore, the contingency management was supervised regarding the use of coercive measures (seclusion and physical restraint) during a weekly meeting of senior psychiatrists, advanced MCP-therapists and therapeutic team leaders. Unfortunately, training data for staff were not collected continuously. No data are available for the joint viewing sessions or for engagement with the online self-study materials. A conservative estimate of the level of reach based on workshop participation indicates that approximately 21% of staff were trained in 2022. This proportion increased to about 47% in 2023 and to approximately 74% in 2024. However, we assume that the actual level of reach is somewhat higher, as the majority of staff were likely reached with at least some modules through the joint viewing sessions conducted at the ward level.

Once basic elements of MCP were established on a ward, problem and functional analyses were implemented as a key element of the initial treatment phase, provided that patients were able to participate based on their symptom burden and cognitive functioning. In this process, patients first describe the central problems that, from their own perspective, led to the inpatient admission. The problematic behavior and its maintaining factors are then analyzed, taking the patient’s perspective into account. Finally, staff and patients collaboratively derive therapy goals and develop a contingency management plan based on the problem and functional analyses. The contingency management plan specifies functional behaviors to be reinforced by the team in order to encourage progress toward the patient’s goals. Importantly, dysfunctional behaviors are not met with sanctions but with the withdrawal of positive reinforcers whenever feasible (for more information see manual; 23).

Measures1

Routine clinical care data

We assessed data on admitted cases, occupied beds, suicide attempts (= number of suicide attempts at site A and B per year), suicides (= number of committed suicides at site A and B per year), assaults (= any form of verbal, nonverbal, or physical behavior that poses a threat to the individual themselves, to others, or to their property), diagnoses (= according to ICD-10), sociodemographic variables, levels of staff training (= number of staff members attending workshops), and coercive measures (see dependent variables). We did not assess data on forced medication, as forced medication was not assessed separately from seclusion and physical restraint.

Dependent variables

All dependent variables were assessed via the documentation systems of the wards, where staff members documented restraints and seclusions and the duration (in hours) of the interventions in detail. We included the categorical dependent variables restraint during treatment (0 = “no”, 1 = “yes”), which reflects 5- and 7-point restraints as well as abdominal belt restraints. Additionally, we included seclusion during treatment (0 = “no”, 1 = “yes”), as well as the continuous dependent variables hours of restraint and hours of seclusion (each log-transformed to account for skewness) on case level. When a case was restrained or secluded multiple times, the total duration of all episodes was summed up. Thus, each case was included only once in the analyses, regardless of how many times coercion occurred. Unfortunately, we were not able to calculate the duration of abdominal belt restraints, as this form of restraint was not assessed in hours. Therefore, the dependent variable hours of restraint included only the duration of 5- and 7-point restraints.

Clinical global impression scale (CGI)

The CGI rating scale relies on subjective impressions of clinicians and taking into account all available, illness-specific information. The scale measures symptom severity, treatment response, and treatment efficacy in individuals with mental disorders, evaluated by clinical professionals [30, 31]. The three subscales include: severity of illness on a seven-point scale from 1 = “normal, not ill at all” to 7 = “among the most extremely ill patients”; improvement during treatment on a seven-point scale from 1 = “very much improved” to 7 = “very much worse”; and the efficacy index with a 4 × 4 rating scale assessing therapeutic effects (including medication) and associated side effects [32]. In the current analyses, mean and median scores of the subscales severity of illness and improvement during treatment were presented.

Although the CGI is widely established, considered as valid, intuitive, and easy to use [32, 33], previous studies have reported mixed results regarding its validity and reliability [34–37]. These findings are partly attributed to ambiguously defined anchors and the high subjectivity of the clinical ratings [33, 38]. To the best of our knowledge, no unified transdiagnostic studies on the validity and reliability of the instrument exist, although it is used extensively in both research and clinical practice.

Global assessment of functioning scale (GAF)

The GAF scale is known worldwide as Axis V of the Diagnostic and Statistical Manual of Mental Disorders 4th Edition (DSM-IV) [39]. The scale is trans-diagnostic and assesses the overall social, occupational and psychological functioning of an individual, rated by clinical professionals [40, 41]. The scale ranges from 1 to 10 = “severely impaired” to 91–100 = “extremely high functioning” (0 = “no information”). As with all observer-rated instruments, the GAF is dependent on the expertise of the clinical rater and is therefore better suited for group-level assessments rather than for drawing conclusions about an individual [42]. Findings regarding its reliability and validity are mixed for different diagnoses and settings [40, 42]. Nonetheless, overall evidence suggests that the GAF possesses satisfactory psychometric properties for the evaluation of the severeness of mental disorders and functioning of life in psychiatric samples [43, 44].

Statistical analyses

We analyzed descriptive group differences between sites A and B using independent t-tests for normally distributed data and Chi-square tests for categorial data as well as Mann-Whitney-U tests for ordinally scaled data. For analyses of age and gender, we excluded cases with missing values (n = 1). For the confirmatory analyses we included all cases.

In a first step, marginal predicted probabilities of experiencing coercive measures (restraint and seclusion) were calculated for each group (site A vs. site B), using the emmeans package (type = “response”; [45]) in R, and pairwise comparisons were conducted to assess overall differences between groups. To analyze H1a - H4b, we employed a series of generalized linear models (GLM) with a logit link function and linear models (LM). We performed all analyses in R [46], using the glm and lm functions of the package stats [47].

For H1a and H1b, we conducted a GLM predicting whether a case was restrained (1) or not (0) during the treatment period with a group dummy (site A = 0, site B = 1), treatment year (2022 = 0, 2023 = 1, 2024 = 2), and their interaction. We included all cases from both sites (n = 15,420). We ran the same model using seclusion (1 = yes, 0 = no) as outcome to test H3a/ H3b.

For H2a and H2b, we conducted a LM with the log-transformed hours of restraint per case as dependent variable, predicted by a group dummy (site A = 0, site B = 1), treatment year (2022 = 0, 2023 = 1, 2024 = 2), and their interaction. Here, we included all cases subjected to coercion (n = 1,015), including cases that were treated multiple times. We ran the same model using the log-transformed hours of seclusion as outcome to test H4a/H4b.

Sample

For the purpose of these analyses, we refer to individual lines of data as treatment cases, as some patients had multiple admissions during the period of data collection (2022–2024) and can therefore appear as multiple cases (repeated admissions occurred in 36% of cases at site A and 35.5% of cases at site B). A total of n = 15,361 cases (n = 10,041 cases at site A and n = 5,320 cases at site B) were analyzed. Mean age across both sites was M = 45.13 (SD = 18.83), with 47.45% of the cases identifying as female and 0.07% as genderqueer. The most common diagnoses were substance use disorders n = 4550 (29.62%), affective disorders n = 3752 (24.43%), and schizophrenia spectrum disorders n = 2392 (15.57%). See Table 1 for detailed information on sample characteristics. Mean GAF values at admission and discharge, as well as CGI severity and improvement scores for the entire sample (2022–2024) at sites A and B are presented in Fig. 2.

Table 1.

Sample characteristics of all cases of the sites A and B (full sample) as well as characteristics of cases subjected to coercive measures (sample cm)

Sample Characteristics	full samplea n/ M (%/ SD)	sample cmb n/ M (%/ SD)
Age	45.13 (18.83)	44.89 (20.31)
Gender
· Female	7288 (47.45)	379 (39.60)
· Male	8062 (52.49)	578 (60.40)
· Genderqueer	10 (00.07)	-
Diagnosis (ICD-10 categories)
· Organic mental disorders (F00-F09)	1345 (8.76)	184 (19.23)
· Substance use disorders (F10-F19)	4550 (29.62)	175 (18.29)
· Schizophrenia, schizotypal and delusional disorders (F20-F29)	2392 (15.57)	352 (36.78)
· Affective Disorders (F30-F39)	3752 (24.43)	112 (11.70)
· Neurotic, stress-related and somatoform (F40-F48)	1619 (10.54)	46 (4.81)
· Behavioral syndromes associated with physiological disturbances and physical factors (F50-F59)	7 (0.05)	-
· Disorders of personality and behavior (F60-F69)	1591 (10.36)	79 (8.25)
· Mental retardation (intellectual disability) (F70-F79)	72 (0.47)	8 (0.84)
· Disorders of psychological development (F80-F89)	18 (0.12)	1 (0.10)
· Behavioral and emotional disorders (onset in childhood and adolescence) (F90-98)	15 (0.10)	-
Global Assessment of Functioning Scalec	*32.83 (11.36)	*21.73 (10.80)
Clinical Global Impression Scaled
· Severity	*5.27 (1.03)	*5.74 (1.32)
· Improvement	*2.17 (1.13)	*1.92 (1.33)

^afull sample n = 15 361, ^bsample cm n = 957

^cGAF: 0 = no information, 1–10 = severely impaired, 91–100 = extremely high functioning

^dCGI: Severity: 1 = normal, not ill at all, 7 = Among the most extremely ill; Improvement: 1 = Very much improved, 7 = very much worse

* p < .000 (t-test or Mann-Whitney-U Test)

Fig. 2.

Mean GAF (= Global Assessment of Functioning Scale) scores across cases at admission and discharge (1–10 = “severely impaired” to 91–100 = “extremely high functioning” (0 = “no information”)). Mean CGI (= Clinical Global Impression Scale) scores across cases for subscales severity (1 = “normal, not ill at all” to 7 = “Among the most extremely ill patients”) and improvement (1 = “very much improved” to 7 = “very much worse”)

Comparing the two sites, with regard to cases subjected to coercion, cases at site A differed from cases at site B on age (M_{age A} = 41.34 (SD_A = 16.59); M_{age B} = 48.15 (SD_B = 22.75); t(909.48) = -5.31, p < .001), as cases at site A were significantly younger. With regard to gender, there was no significant difference between sites (site A: n = 183 (39.96%) women, site B: n = 196 (39.27%) women; χ2(1) = 0.023, p = .882). Looking at the GAF scores, cases at site B were rated as significantly more ill at admission (M_GAF-A = 22.96 (SD_A = 10.84); M_GAF-B = 20.6 (SD_B = 10.65); t(920.32) = 3.35, p = .001), even though cases at both sites fell within the same severe GAF category (“behavior is considerably influenced by delusions or hallucinations or serious impairment, in communication or judgment, e.g., sometimes incoherent, acts grossly inappropriately, suicidal preoccupation, or inability to function in almost all areas, e.g., stays in bed all day, no job, home, or friends”). CGI severity scores did not differ between sites (M_{CGIsev A}= 5.8, (SD_A = 1.12), Median_A = 6; M_{CGIsev B}= 5.68, (SD_B = 1.48), Median_B = 6; W = 109921, p = .924).

Results

Descriptive results for outcome and context variables

Out of all assessed cases, n = 957 (6.23%) individual cases were subjected to coercive measures during their treatment n = 458 at site A (4.56% of cases at site A) and n = 499 at site B (9.38% of cases at site B). Across sites, the proportion of men was higher among those who were subjected to coercive measures (60.40%) than in the total sample (52.49%). Additionally, the diagnoses of cases subjected to coercive measures differed from the most common diagnoses in the overall sample (see Table 1). Among these cases, 36.78% (n = 352) met criteria for schizophrenia spectrum disorder, 19.23% (n = 184) for an organic mental disorder (i.e., dementia), and 18.29% (n = 175) for a substance used disorder. Thus, those with schizophrenia spectrum disorder and organic mental disorder were overrepresented, compared to the full sample. In addition to this, cases subjected to coercive measures were rated as significantly more ill (see Table 1) when entering the psychiatric hospital, compared to those cases who ultimately did not experience coercive measures. At both sites, cases who were subjected to coercive measures during the treatment period were categorized as “markedly ill” to “severely ill” (see Table 1).

Table 2 presents descriptive data for coercive measures and context variables by assessment year. Figure 3 shows the trajectories of percentage values of the dependent variables “restraint / seclusion during treatment” at site A and B and the “average hours of restraint/ seclusion per case” at site A and B from 2022 to 2024. The estimated probability of experiencing physical restraint was 2.9% (95% CI: 2.6–3.3%) at site A and 6.6% (95% CI: 5.9–7.3%) at site B. The between-group comparison indicated a significant difference (OR = 0.42, p < .001), showing that overall, the likelihood of being restraint was substantially higher at site B than at site (A) We found an equivalent pattern for seclusion. Here, the estimated probability of experiencing isolation was 4.4% (95% CI: 3.9–4.8%) at site A and 7.2% (95% CI: 6.6–7.8%) at site (B) The between-group comparison indicated a significant difference (difference = − 2.9% points, p < .001), showing that overall, the likelihood of isolation was substantially higher at site B compared to site A.

Fig. 3.

Trajectories of percentage values of restraint and seclusion from 2022–2024 and average duration of restraint and seclusion per treatment case from 2022–2024

In addition to that, we found that the use of observation rooms on the wards at site A declined over the years. In 2022, observation rooms often reached capacity limits at site A (see Fig. 4) leading to severe care problems in daily psychiatric practice. In 2024, one observation room was permanently closed and, on average, three to four out of the eight remaining observation rooms were empty (unfortunately, we did not have access to equivalent data from site B).

Fig. 4.

Degree of capacity utilization of the observation rooms at site A from 2022–2024

In addition to this, Table 2 presents data for relevant context variables that are associated with an increased likelihood of coercive measures in literature [48, 49]. Due to the small number of cases compared to the overall sample, inferential statistical analyses of these variables were not performed. The number of assaults descriptively declined by 50% at site A from 2022 to 2024, whereas they remained descriptively unchanged at site B. Suicide attempts descriptively increased from 2022 to 2024 at site B from n = 2 to n = 11 but remained nearly unchanged at site A (n = 4 at 2022 and n = 3 at 2024), whereas the rate of committed suicides remained constant at both sites.

Confirmatory analyses

Restraint

First, we performed a GLM to test hypotheses H1a and H1b. Results indicate that the likelihood of being restraint during the treatment period significantly declined from 2022 to 2024 at site A (Est. = -0.22, SE = 0.07, p = .004, OR = 0.81, CI[0.70, 0.93]), supporting hypothesis H1a. The interaction term year × group failed to reach statistical significance (Est. = 0.06, SE = 0.10, p = .531, OR = 1.07, CI[0.87, 1.30]), indicating that decline did not differ significantly between sites. Thus, hypothesis H1b was not supported. In the literature, some variables such as male gender, more severe mental illness or younger age were associated with a higher probability for being subjected to coercive measures [8, 50, 51]. To adjust for those possible confounders, we recalculated the unadjusted model including age, gender and severity of the mental illness (CGI severity score) as covariates. The inclusion of covariates did not change the results of the model. Male gender and higher values on the CGI severity scale were significantly associated with a higher probability of being subjected to restraint (see Table 3 for an overview of the adjusted models). In a second step, we analyzed whether the time of restraint (in hours) declined at site A from 2022 to 2024 (H2a), using a LM. Furthermore, we assessed whether the decline at site A was stronger in comparison to site B from 2022 to 2024 (H2b). Results of the linear model indicate that the duration of restraint significantly declined at site A (Est. = -0.27, SE = 0.10, p = .005, exp(b) = 0.76, CI[0.63,0.92]) supporting hypothesis H2a. Nevertheless, the non-significant interaction term shows that the difference between the groups is not statistically meaningful (Est. = 0.05, SE = 0.13, p = .703, exp(b) = 1.05, CI[0.81,1.36]), indicating that H2b was not supported.

Table 3.

Adjusted models for physical restraint and seclusion

	b	Restraint					b	Seclusion
		SE		OR	CI	p		SE		OR	CI	p
Intercept	-7.61	0.35		0.00	0.00–0.00	< 0.001	-0.01	0.01		0.94	0.92–0.96	< 0.001
Year	-0.26	0.08		0.77	0.66–0.89	0.001	-0.00	0.01		1.00	0.99-1.00	0.672
Site	0.67	0.12		1.96	1.53–2.51	< 0.001	0.02	0.00		1.02	1.01–1.03	< 0.001
Age	-0.00	0.00		0.99	0.99-1.00	0.223	-0.00	0.00		0.99	0.99-1.00	< 0.001
Gender	0.35	0.09		1.42	1.21–1.69	< 0.001	0.02	0.00		1.01	1.01–1.02	< 0.001
CGI_severity	0.76	0.06		2.14	1.91–2.40	< 0.001	0.02	0.00		1.02	1.01–1.03	< 0.001
Site*Year	0.16	0.10		1.17	0.95–1.44	0.129	0.00	0.01		1.01	0.99–1.02	0.133

Note. Site was dummy coded as 0 = site A (with MCP), 1 = site B (with TAU). Year was coded as 0 = 2022, 1 = 2023, 2 = 2024, Gender: 0 = female, 1 = male, 2 = divers, CGI_severity = 1–7. OR = odds ratio. CI = 95% confidence interval for the odds ratio

Seclusion

To assess whether the likelihood of being secluded during the treatment period declined from 2022 to 2024 after implementation of MCP at site A (H3a) and whether this decline outperformed a possible decline from 2022 to 2024 at site B (H3b) we performed another GLM. Results showed that the likelihood of being secluded during treatment remained relatively stable over time (2022–2024) at site A (Est. = -0.00, SE = 0.00, p = .893, OR = 1.00, CI[0.99,1.01]), indicating that H3a was not supported. The non-significant interaction term (site x treatment year) shows that the trajectories between the two sites were comparable stable over the years 2022–2024 (Est. = 0.00, SE = 0.00, p = .829, OR = 1.00, CI[0.99,1.01]), indicating no support for hypothesis H3b. We also recalculated the GML for seclusion, adjusting for possible confounders. The inclusion of covariates (gender, age and CGI severity scale) did not change the results of the model. Alle three covariates male gender, younger age and higher values of the CGI severity scale were significantly associated with seclusion (see Table 3 for an overview of the adjusted models).

To test H4a and H4b we conducted another LM. We assessed whether the duration of seclusion (in hours) significantly declines with the implementation of MCP from 2022 to 2024 at site A (H4a). Results show that the time of seclusion significantly declined at site A (Est. = -0.40, SE = 0.14, p = .006, exp(b) = 0.67, CI[0.50,0.90]), supporting hypothesis H4a. The significant interaction term between treatment site and implementation year indicates that the decline at site A is significantly stronger compared to site B (Est. = 0.534, SE = 0.21, p = .01, exp(b) = 1.71, CI[1.12,2.60]), supporting hypothesis H4b.

Discussion

This naturalistic, multi-year evaluation examined whether implementing Modular Competence-Oriented Psychotherapy (MCP) at one inpatient psychiatric site was associated with changes in the use of coercive measures compared to a second site continuing treatment as usual. Overall, the implementation of MCP at site A appeared feasible and clinically promising. Reductions in coercive measures occurred without increases in other crisis indicators such as suicide attempts or assaults, and these findings align with prior feasibility reports from other clinical settings [28, 52]. Given the large sample of routine clinical data spanning three years (n = 15,361 cases with n = 957 being subjected to coercive measures), the results offer ecologically valid insight into the potential contribution of a competence-oriented psychotherapeutic framework to reducing coercive practices in routine psychiatric care.

Psychotherapy studies conducted under real-world clinical conditions remain uncommon despite longstanding calls for such work [53–55]. The present study is among the few that address the implementation of psychotherapy in routine psychiatric care with a specific focus on the development of coercive measures and assaults. This emphasizes a high ecological validity of the results, with regard to the use of MCP in general psychiatric care. In detail, the retrospective analysis of the available data provides preliminary evidence suggesting that the implementation of MCP could have a positive impact on various aspects of routine psychiatric care.

One particular aspect is probably the use of coercive measures. Overall, the comparison of the model-based averages between the sites shows, that the probability for cases to experience restraint and/ or seclusion is significantly higher at site B in comparison to site A over the years 2022–2024. These results provide first preliminary evidence that the implementation of MCP may be superior to routine psychiatric care. In addition to that, data from site A showed a significant reduction in the number of physical restraints, the duration of restraints, and the duration of seclusion under the implementation of MCP (support for the hypotheses H1a, H2a and H4a). For duration of seclusion, the decline at site A was significantly stronger than changes at site B with TAU, supporting hypothesis H4b. Descriptively, the data also indicated that the number of assaults by patients - whether against staff, themselves, or other patients - was reduced by approximately half and suicide attempts remained unchanged. Given that MCP is designed to target dysfunctional behavior, the findings could be consistent with the anticipated effect of its implementation. However, systematic randomized controlled trails (RCT) are necessary to establish causality in this association (see also limitation section). The observed improvements also mirror effects reported for coercion-focused interventions such as Safe Wards or Six Core Functions [8], while MCP offers a broader therapeutic framework that extends beyond crisis management to support ongoing, cross-professional intervention consistency.

Another clinically important effect is that the use of the observation rooms markedly declined, which could be associated with the reduced application and shorter duration of restraint and seclusion. Empty observation rooms in psychiatric care contribute to greater predictability and safety for staff, as they ensure the availability of a protected space during escalating situations. In the long term, this may lead to increased employee satisfaction and reduced psychological stress among staff members. Unfortunately, we did not have data of the use of observational rooms at site B, and were thus unable to compare the two sites in terms of statistical measures.

Looking at site B where MCP was not implemented, results indicate that physical restraints (number and duration) also decline at site B from 2022 to 2024 (see also Fig. 3). Furthermore, the decline in number of restraints and duration of restraints at site A with MCP was not significantly different from those at site B (no support for hypotheses H1b and H2b). One possible interpretation of these findings could be the occurrence of an observer effect (also knwon as “hawthorne effect”; 16, 29), as staff at site B were aware that MCP was being introduced at site A since the beginning of 2024. Furthermore, there was an increased critical discussion initiated by the management and administration with the staff concerning the use of coercive measures since the beginning of 2024 at site B, as the first descriptive data of site A were discussed at leadership meetings at site B. In addition to this, results show an increase in the number and duration of seclusion at site B, on a descriptive level (see Fig. 3), which could indicate a compensatory shift in coercive strategies. One plausible interpretation is that patients were subjected to longer seclusion periods to avoid the need for restraint. In addition, suicide attempts increased markedly between 2022 and 2024 at site B, which may indicate that existing safety structures deteriorated without the implementation of adequate compensatory interventions to deal with crisis situations. Regarding violent incidents, the number of assaults at site B remained stable over the observation period, whereas it declined at site A.

Taken together, the simultaneous implementation of MCP and reduction in coercive measures provide initial evidence that MCP could be suited to address dysfunctional aggressive behavior, as do other interventions with the aim to reduce coercive measures [11, 12, 16], but within a broader therapeutic framework. Beyond the reduction of coercive measures, the implementation of feasible therapeutic frameworks in psychiatric care conforms to the claimed standards in modern psychiatry [22]. In this respect, MCP promises to be a suitable framework to meet the demand for personalized, trans-diagnostic therapy [56], including the improvement of competence and skills for the whole psychiatric team and thereby reducing one core ethical problem in modern psychiatry: the use of coercive measures.

Limitations

In addition to the promising results mentioned above, the design of the current retrospective analysis has some limitations. First, the results are based on data from routine clinical practice that were retrospectively evaluated. On the one hand, this provides high ecological validity and demonstrates that MCP can be implemented in routine psychiatric care. Furthermore, using ecological valid data from routine care allows analyzing longer time periods than intervals during RCTs and it is possible to include all cases without being restricted to inclusion and exclusion criteria as demanded for RCTs, which can lead to highly selective samples [3]. On the other hand, the fact that the current analyses are not based on an RCT includes the risk for confounding effects, which may be caused by site-specific conditions, staff turnover, or other confounding variables [57]. Furthermore, a RCT could provide evidence of causality regarding the effects of MCP implementation, which the current investigation cannot. Thus, running a study based on an RCT-design is needed to replicate the current findings. In addition to this, prospectively planed studies should also include possible co-variates like age or gender, which were not accounted for in the current analyses.

Second, it could be possible that structural changes, beyond the implementation of MCP, are responsible for the current findings, such as patients exhibiting lower psychopathology levels at site A in the years 2023 and 2024 in comparison to site B. Given that our data were retrospectively analyzed, naturalistic data, we were not able to regulate the admissions in psychiatry in terms of severity of the mental illness or diagnoses. Nevertheless, comparable GAF and CGI severity ratings show that cases at both sites stayed comparable throughout the period under review.

Third, the data were only collected at two sites (A and B of the Vitos Clinic Giessen-Marburg). Future training in MCP should also be systematically conducted at independent clinics to more reliably measure the degree of MCP implementation. Additionally, it should be noted that the degree of MCP implementation could currently only be determined through feedback from the wards, as there is no adherence scale for MCP implementation at present. In order to evaluate the effectiveness of MCP in reducing coercive measures, mandatory supervision and analysis of executed coercive measures in a structured setting would also need to be performed. Thus, the current analyses can only provide initial evidence that the introduction of MCP could be related to the reduction in coercive measures at site A.

Conclusion and further directions

The present findings indicate first evidence that MCP can be implemented successfully in routine inpatient psychiatric care, which supports other findings of implementing MCP in outpatient care [52]. At the site where MCP was introduced, we observed consistent reductions in the number and duration of restraints and in the duration of seclusion, without evidence of increased assaults or suicide attempts. These changes emerged under real-world conditions, across all wards, and in a large, diagnostically heterogeneous patient group, underscoring the potential of a competence-oriented psychotherapeutic framework for all professions to improve everyday clinical practice.

However, due to the retrospective design of the analyses and possible observer effects at site B, we do not claim causality for an association between the implementation of MCP and the reduction in coercive measures. To reevaluate the current findings, prospectively designed and controlled studies are necessary. To build on our first results from inpatient and outpatient care [52] and to address some limitations of the current naturalistic design, staff at site B have now begun MCP training supported by a newly developed adherence scale, staff surveys about the usability of MCP, structured documentation of coercive events, as well as supervision for executed coercive measures. These steps are intended to enhance implementation fidelity and allow more rigorous evaluation of the effects of MCP on clinical treatment.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

LMS drafted the manuscript and performed the data analyses. JH contributed substantially to writing and editing the manuscript and supervised the analyses. YB edited the manuscript and provided critical input regarding the analyses. MF contributed substantially to the design of the modular competence-oriented psychotherapy and to the conception of the study design; he also edited the manuscript. SF contributed substantially to the design of the modular competence-oriented psychotherapy and edited the final manuscript. All authors read and approved the final manuscript.

Funding

Not applicable.

Data availability

The datasets analyzed during the current study are not publicly available. The manuscript is based on a retrospective data analysis of routine data from two psychiatric sites. Although the project itself is covered by ethics approval, consent from the patients was not obtained for uploading data to a server. Therefore, under EU law, it is not possible to upload clinical data without prior consent from the patients, as we were not able to guarantee that indirect identification is impossible.

Declarations

Ethics approval and consent to participate

These retrospective analyses are part of an evaluation study on the effectiveness and feasibility of MCP in collaboration with the University of Leipzig (ethics approval was granted by Landesaerztekammer Hessen; LAEKH: 2022-2943-evBO, in accordance with the Declaration of Helsinki). The data of the current analyses are retrospective, routine data from two psychiatric sites. According to national regulations (§ 27 Bundesdatenschutzgesetz (BDSG; German Federal Data Protection Act)) in conjunction with Article 9(2)(j) of the General Data Protection Regulation (GDPR; EU-Regulation), informed consent of the participants was not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.