Researchers suggested that digital twins could eventually support short-term violence risk, treatment scenario modeling, and discharge preparation in forensic mental health, while cautioning that the technology may be speculative and ethically complex.
In a narrative review, the researchers examined emerging literature on digital twins in health care, digital psychiatry, forensic mental health risk management, and related ethical frameworks. They synthesized peer-reviewed studies, consensus statements, and policy documents to map potential applications, technical constraints, and governance requirements in forensic mental health.
Forensic mental health services provide care to patients with serious mental disorders who have offended or may pose substantial risk, often across prisons, courts, secure hospitals, and community settings. Current structured professional judgment tools, including the Historical Clinical Risk Management 20, Violence Risk Appraisal Guide, and Psychopathy Checklist Revised provide systematic risk assessment frameworks but are limited by point-in-time evaluations, variable predictive validity, and reduced relevance over time.
Digital twins are dynamic computational models that integrate multiple data streams and update as new information becomes available. The researchers proposed a three-level digital twin framework in forensic mental health comprising person-level models for dynamic risk estimation and care planning; ward-level models for operational and environmental scenario testing; and pathway-level models for patient flow, resource planning, and policy evaluation.
The researchers didn't present pooled quantitative analyses or subgroup findings. They highlighted that the evidence base remains conceptual, with no published studies describing fully implemented and validated digital twins in routine forensic mental health practice.
Potential person-level data streams included electronic health records, medication history, psychological assessments, incident logs, legal documentation, wearable-derived sleep and activity data, proximity sensors, environmental measures such as noise or crowding, and patient-reported outcomes. In theory, these models could provide short-term forecasts over hours to days for violence, self-harm, or absconding risk. However, the researchers noted that algorithms capable of reliably distinguishing imminent risk from baseline variation in forensic populations haven't been validated.
Digital phenotyping was described as one enabling technology. Smartphone and wearable data have been associated with depression and anxiety severity, psychosis, bipolar disorder, sleep regularity, mood stability, and relapse risk in nonforensic psychiatric populations. The researchers cautioned that these associations may not generalize to secure settings, where involuntary detention, institutional routines, and restricted movement could alter baseline behavioral and physiologic patterns.
The researchers also described potential ward-level applications, including scenario testing for the admission of high-acuity patients, staff scheduling, de-escalation training, and physical environment redesign. They characterized violence risk assessment, medication optimization, leave progression, and conditional discharge as speculative or highly speculative use cases, with seclusion and restraint reduction considered only partially feasible because some incident-pattern analysis can already be performed through existing quality improvement methods.
The researchers proposed a staged implementation pathway over the course of more than 5 years. The first phase would focus on foundational research, technical feasibility, ethical framework development, stakeholder engagement, and proof of concept. Later phases included single-site pilots, staff training, user feedback, safety validation, multisite deployment, outcome evaluation, governance refinement, performance monitoring, system-wide adoption with continuous monitoring, and policy integration.
Ethical and legal concerns were central to the review. The researchers warned that continuous monitoring in forensic settings could intensify coercion, normalize surveillance, erode privacy, and encourage patients to modify behavior to appear lower risk. They also noted that models trained on historical forensic data could perpetuate racial, gender, or socioeconomic inequities if past clinical and institutional decisions reflected those biases.
The researchers recommended safeguards, including human rights impact assessment, data minimization, access controls, audit trails, bias monitoring, explainable models, patient advocacy involvement, and clinician oversight. They advised against fully automated decision-making, long-term outcome prediction beyond 6 months for discharge decisions, deployment in settings without independent tribunal review or legal advocacy, and use of digital twins primarily for security or custodial purposes.
Limitations of the review included the absence of systematic search and meta-analytic methods, lack of formal quality assessment, reliance on adjacent fields, rapidly changing technology and regulation, international variation in forensic mental health systems, and limited attention to patient and carer perspectives.
“Current evidence does not support widespread deployment of digital twins in forensic mental health,” wrote lead study author James O. Olawade, of the Department of Guidance and Counselling at the Adekunle Ajasin University in Nigeria, and colleagues. “Enthusiasm for innovation must be tempered by rigorous skepticism, especially when the stakes involve liberty, dignity, and fundamental rights.”
The study authors reported no competing interests.