Regional Self-Harm Strategy Linked to Sharp Drop in Coercive Psychiatric Interventions, With Staff Attitudes Improving After Training

A system-wide self-harm strategy introduced in a Danish region was followed by an immediate and sustained reduction in coercive interventions in psychiatric inpatient care, according to a quasi-experimental study published in Borderline Personality Disorder and Emotion Dysregulation. Annual coercive episodes across five adult inpatient units fell from 1,280 in 2021 to 409 in 2024, while staff attitudes toward patients who self-harm improved after training and remained more positive eight months later.

The intervention was the Regional Self-Harm Strategy announced by Mental Health Services in Region Zealand in June 2022, a multi-component framework emphasizing validation, collaborative crisis planning, and structured de-escalation. It drew on the Safewards model of relational safety, the recovery-oriented Tidal Model, and Dialectical Behavior Therapy-informed principles of emotion regulation. Structured staff training, a five-hour curriculum designed with input from a person with lived experience of severe self-harm, was rolled out to all inpatient staff between May and June 2023. Each unit appointed two local trainers who supported implementation through supervision, consultation, and case-based reflection, and all five units completed the training as planned.

The study used a dual quasi-experimental design. First, the researchers performed an interrupted time-series analysis of monthly registry counts of coercive interventions, mechanical restraint with belts, straps, or holding, and rapid tranquillization, from January 2021 to December 2024. Second, they assessed staff attitudes with a three-wave pre-post survey using the revised 17-item Danish Self-Harm Antipathy Scale at baseline, two months, and eight months after training. Two milestones were pre-specified: June 2022 for the strategy announcement and June 2023 for training implementation. Because the strategy was implemented system-wide, randomization was infeasible, and no contemporaneous control region was available.

The interrupted time-series results showed a large immediate reduction in coercive interventions at the strategy announcement, with an incidence-rate ratio of 0.39 (95% confidence interval 0.21 to 0.73), and no subsequent change in trend. At the training point, a temporary increase appeared, with an incidence-rate ratio of 2.58 (95% CI 1.57 to 4.25), which the authors interpret as short-term implementation disruption rather than an effect of the training itself. Mean monthly episodes fell from 93 before the strategy to 28 during the policy phase, then rose to 38 after training, stabilizing at the lower level. The overall pattern reflected two step changes rather than gradual trends.

Reductions were not uniform across coercion types. Mechanical restraint showed the largest declines: between 2021 and 2024, use of belts decreased by 74 percent, straps by 81 percent, and holding by 82 percent. Rapid tranquillization decreased by 51 percent but remained the most frequently used coercive measure, a substitution pattern the authors say warrants monitoring to ensure one form of containment is not simply replacing another. The composition of coercion also shifted by patient group: in 2021, self-harming patients accounted for most episodes, 960 versus 320 among non-self-harming patients, but by 2023 and 2024 episodes involving non-self-harming patients exceeded those involving self-harming patients.

Sensitivity analyses addressed the concern that a small number of frequently restrained patients drove the decline. Across all years, the top 10 percent of patients accounted for 52 to 58 percent of episodes and the top 5 percent for 35 to 41 percent, proportions that remained stable. Removing episodes from the top 10 percent changed annual totals by less than five percent, and an analysis capping each patient at one episode per year showed a similar trajectory. The authors conclude the decline reflected a broad change in clinical practice rather than shifts among high-frequency patients.

Staff attitudes also changed. Of 187 staff who attended training, 182 completed the baseline questionnaire, and 113 provided valid attitude data at two or more time points. Mean Self-Harm Antipathy Scale scores decreased from 40.0 at baseline to 34.8 at two months and 35.4 at eight months, with lower scores indicating more positive attitudes. The largest raw-score improvement appeared on the Judgmental Perception subscale, reflecting reduced morally judgmental and blame-oriented perceptions. Among participants with matched data, scores declined significantly from baseline to two months and from baseline to eight months, with no further change between the follow-up points.

The authors note an important caveat about the attitude data. The cross-sectional means reflect both within-person change and differential attrition, as participants with initially higher antipathy scores were more likely to drop out. Within-person estimates from paired comparisons and mixed-effects models showed more moderate improvements of approximately three points, which the researchers treat as the less biased figures. Attrition was highest at eight months and reflected shift schedules, staff turnover, and survey fatigue in high-intensity inpatient settings.

An intriguing aspect of the findings is their timing. The reduction in coercive interventions preceded the measured improvement in attitudes: coercion fell sharply after the 2022 strategy announcement, whereas attitude change was assessed in relation to the 2023 training. The authors argue this sequence is difficult to reconcile with social-cognitive models in which attitude change precedes and drives behavioral change. They suggest organizational and symbolic change, leadership commitment, shared framing, and increased monitoring can initiate behavioral shifts before individual attitudes formally change, with attitudinal improvement then serving a consolidating function. The temporary rise during training may represent a defensive rebound, a short return to familiar control practices as new relational routines were rehearsed under real-world constraints.

The study has substantial limitations that the authors discuss openly. The absence of randomization and of a control series leaves the analyses vulnerable to unmeasured influences and concurrent initiatives, and the researchers did not benchmark changes against regional or national coercion trends, so they cannot determine how much of the step change reflects strategy-specific effects versus broader secular developments. Incomplete denominators such as patient-days limited the analysis to counts, meaning changes in bed occupancy, admission patterns, or case-mix could have contributed. Patients were also classified as self-harming or not at the patient level rather than the episode level, so some coercive episodes involving self-harming patients may have been triggered by aggression toward others. Despite these constraints, the pre-specified milestones, long observation window, and robustness checks support the interpretation that the strategy and training were linked to real behavioral change, and the authors emphasize that sustained reductions likely require continued organizational and relational supports beyond training alone.

Subject of Research: Quasi-experimental evaluation of a regional self-harm strategy and staff training on coercive interventions and clinician attitudes in Danish psychiatric inpatient units

Article Title: Does changing clinicians’ attitudes toward self-harm reduce coercive interventions in mental health inpatient settings? A quasi-experimental study of staff training

Article References: Høst Poulsen, K., Hjorthøj, C., Bach, B., Simonsen, E., & Kongerslev, M. T. (2026). Does changing clinicians’ attitudes toward self-harm reduce coercive interventions in mental health inpatient settings? A quasi-experimental study of staff training. Borderline Personality Disorder and Emotion Dysregulation, 13(1), Article 25. https://doi.org/10.1186/s40479-026-00346-2

Image Credits: AI Generated

DOI: 10.1186/s40479-026-00346-2

Keywords: self-harm, coercive interventions, psychiatric inpatient care, staff training, interrupted time-series, mechanical restraint, rapid tranquillization, staff attitudes, Region Zealand, implementation science, quasi-experimental study, de-escalation