Protocol of an AI-based chatbot as an adjunctive tool for reducing psychiatric emergency visits in borderline personality disorder

Abstract

Background: Borderline personality disorder (BPD) is one of the most severe and challenging psychiatric conditions, characterized by emotional dysregulation, impulsivity and instability in selfimage and interpersonal relationships. Despite a community prevalence of approximately 1.5%, BPD accounts for up to 22% of psychiatric inpatients and between 9% and 25% of all psychiatric emergency visits, reflecting high rates of comorbidity, psychotic-like symptoms, self-harm and suicidal behaviour. Around 75% will attempt suicide at least once, and of 2-5% die by suicide. Although psychotherapeutic approaches such as Dialectical Behavioural Therapy (DBT) and Mentalization-Based Treatment (MBT) have proven efficacy, access to specialized management and adherence to long-term care remain limited. Consequently, patients often rely on emergency services as a maladaptive coping mechanism, increasing distress, functional impairment, disability and healthcare costs, while perpetuating societal, institutional and self-stigma. In Catalonia, the average annual cost per patient is 11,308€ and the baseline of total BPD cost is 45.6 million€; only 35% related to mental health care. Even though BPD remains underdiagnosed, undertreated and underinvestigated. Objectives: This study aims to evaluate whether the implementation of a protocolized artificial intelligence (AI) chatbot intervention can serve as an adjunctive clinical tool to standard pharmacological and/or psychological treatment for patients with BPD, reducing psychiatric emergencies and enhancing emotional regulation, psychoeducation and patient autonomy, while improving adherence and satisfaction with care. Methods: A multicentre randomized controlled clinical trial will be conducted in the seven adult mental health centres (CSMAs) in the province of Girona. The study will enrol adults aged 18-45 diagnosed with BPD according to DSM-5-TR criteria. Eligible participants will be randomized (1:1) using a computer-generated sequence to either the control (treatment as usual) or the experimental group (treatment as usual plus access to the AI chatbot). The chatbot is designed and trained to provide psychoeducation, emotional support and crisis vigilance and management strategies based on DBT and MBT frameworks. Allocation and data collection will be managed through REDCap’s electronic case report forms

3