Defining acceptable data collection and reuse standards for queer artificial intelligence research in mental health: protocol for the online PARQAIR-MH Delphi study.

INTRODUCTION: For artificial intelligence (AI) to help improve mental healthcare, the design of data-driven technologies needs to be fair, safe, and inclusive. Participatory design can play a critical role in empowering marginalised communities to take an active role in constructing research agendas and outputs. Given the unmet needs of the LGBTQI+ (Lesbian, Gay, Bisexual, Transgender, Queer and Intersex) community in mental healthcare, there is a pressing need for participatory research to include a range of diverse queer perspectives on issues of data collection and use (in routine clinical care as well as for research) as well as AI design. Here we propose a protocol for a Delphi consensus process for the development of PARticipatory Queer AI Research for Mental Health (PARQAIR-MH) practices, aimed at informing digital health practices and policy. METHODS AND ANALYSIS: The development of PARQAIR-MH is comprised of four stages. In stage 1, a review of recent literature and fact-finding consultation with stakeholder organisations will be conducted to define a terms-of-reference for stage 2, the Delphi process. Our Delphi process consists of three rounds, where the first two rounds will iterate and identify items to be included in the final Delphi survey for consensus ratings. Stage 3 consists of consensus meetings to review and aggregate the Delphi survey responses, leading to stage 4 where we will produce a reusable toolkit to facilitate participatory development of future bespoke LGBTQI+-adapted data collection, harmonisation, and use for data-driven AI applications specifically in mental healthcare settings. ETHICS AND DISSEMINATION: PARQAIR-MH aims to deliver a toolkit that will help to ensure that the specific needs of LGBTQI+ communities are accounted for in mental health applications of data-driven technologies. The study is expected to run from June 2024 through January 2025, with the final outputs delivered in mid-2025. Participants in the Delphi process will be recruited by snowball and opportunistic sampling via professional networks and social media (but not by direct approach to healthcare service users, patients, specific clinical services, or via clinicians' caseloads). Participants will not be required to share personal narratives and experiences of healthcare or treatment for any condition. Before agreeing to participate, people will be given information about the issues considered to be in-scope for the Delphi (eg, developing best practices and methods for collecting and harmonising sensitive characteristics data; developing guidelines for data use/reuse) alongside specific risks of unintended harm from participating that can be reasonably anticipated. Outputs will be made available in open-access peer-reviewed publications, blogs, social media, and on a dedicated project website for future reuse.

Management of antipsychotics in primary care: Insights from healthcare professionals and policy makers in the United Kingdom.

INTRODUCTION: Antipsychotic medication is increasingly prescribed to patients with serious mental illness. Patients with serious mental illness often have cardiovascular and metabolic comorbidities, and antipsychotics independently increase the risk of cardiometabolic disease. Despite this, many patients prescribed antipsychotics are discharged to primary care without planned psychiatric review. We explore perceptions of healthcare professionals and managers/directors of policy regarding reasons for increasing prevalence and management of antipsychotics in primary care. METHODS: Qualitative study using semi-structured interviews with 11 general practitioners (GPs), 8 psychiatrists, and 11 managers/directors of policy in the United Kingdom. Data was analysed using thematic analysis. RESULTS: Respondents reported competency gaps that impaired ability to manage patients prescribed antipsychotic medications, arising from inadequate postgraduate training and professional development. GPs lacked confidence to manage antipsychotic medications alone; psychiatrists lacked skills to address cardiometabolic risks and did not perceive this as their role. Communication barriers, lack of integrated care records, limited psychology provision, lowered expectation towards patients with serious mental illness by professionals, and pressure to discharge from hospital resulted in patients in primary care becoming 'trapped' on antipsychotics, inhibiting opportunities to deprescribe. Organisational and contractual barriers between services exacerbate this risk, with socioeconomic deprivation and lack of access to non-pharmacological interventions driving overprescribing. Professionals voiced fears of censure if a catastrophic event occurred after stopping an antipsychotic. Facilitators to overcome these barriers were suggested. CONCLUSIONS: People prescribed antipsychotics experience a fragmented health system and suboptimal care. Several interventions could be taken to improve care for this population, but inadequate availability of non-pharmacological interventions and socioeconomic factors increasing mental distress need policy change to improve outcomes. The role of professionals' fear of medicolegal or regulatory censure inhibiting antipsychotic deprescribing was a new finding in this study.

Clinical Prompt Learning With Frozen Language Models.

When the first transformer-based language models were published in the late 2010s, pretraining with general text and then fine-tuning the model on a task-specific dataset often achieved the state-of-the-art performance. However, more recent work suggests that for some tasks, directly prompting the pretrained model matches or surpasses fine-tuning in performance with few or no model parameter updates required. The use of prompts with language models for natural language processing (NLP) tasks is known as prompt learning. We investigated the viability of prompt learning on clinically meaningful decision tasks and directly compared this with more traditional fine-tuning methods. Results show that prompt learning methods were able to match or surpass the performance of traditional fine-tuning with up to 1000 times fewer trainable parameters, less training time, less training data, and lower computation resource requirements. We argue that these characteristics make prompt learning a very desirable alternative to traditional fine-tuning for clinical tasks, where the computational resources of public health providers are limited, and where data can often not be made available or not be used for fine-tuning due to patient privacy concerns. The complementary code to reproduce the experiments presented in this work can be found at https://github.com/NtaylorOX/Public_Clinical_Prompt.

Automatic Detection of Cognitive Impairment with Virtual Reality.

Cognitive impairment features in neuropsychiatric conditions and when undiagnosed can have a severe impact on the affected individual's safety and ability to perform daily tasks. Virtual Reality (VR) systems are increasingly being explored for the recognition, diagnosis and treatment of cognitive impairment. In this paper, we describe novel VR-derived measures of cognitive performance and show their correspondence with clinically-validated cognitive performance measures. We use an immersive VR environment called VStore where participants complete a simulated supermarket shopping task. People with psychosis (k=26) and non-patient controls (k=128) participated in the study, spanning ages 20-79 years. The individuals were split into two cohorts, a homogeneous non-patient cohort (k=99 non-patient participants) and a heterogeneous cohort (k=26 patients, k=29 non-patient participants). Participants' spatio-temporal behaviour in VStore is used to extract four features, namely, route optimality score, proportional distance score, execution error score, and hesitation score using the Traveling Salesman Problem and explore-exploit decision mathematics. These extracted features are mapped to seven validated cognitive performance scores, via linear regression models. The most statistically important feature is found to be the hesitation score. When combined with the remaining extracted features, the multiple linear regression model resulted in statistically significant results with R2 = 0.369, F-Stat = 7.158, p(F-Stat) = 0.000128.

Explainable artificial intelligence for mental health through transparency and interpretability for understandability.

The literature on artificial intelligence (AI) or machine learning (ML) in mental health and psychiatry lacks consensus on what "explainability" means. In the more general XAI (eXplainable AI) literature, there has been some convergence on explainability meaning model-agnostic techniques that augment a complex model (with internal mechanics intractable for human understanding) with a simpler model argued to deliver results that humans can comprehend. Given the differing usage and intended meaning of the term "explainability" in AI and ML, we propose instead to approximate model/algorithm explainability by understandability defined as a function of transparency and interpretability. These concepts are easier to articulate, to "ground" in our understanding of how algorithms and models operate and are used more consistently in the literature. We describe the TIFU (Transparency and Interpretability For Understandability) framework and examine how this applies to the landscape of AI/ML in mental health research. We argue that the need for understandablity is heightened in psychiatry because data describing the syndromes, outcomes, disorders and signs/symptoms possess probabilistic relationships to each other-as do the tentative aetiologies and multifactorial social- and psychological-determinants of disorders. If we develop and deploy AI/ML models, ensuring human understandability of the inputs, processes and outputs of these models is essential to develop trustworthy systems fit for deployment.

Psychosocial markers of age at onset in bipolar disorder: a machine learning approach.

BACKGROUND: Bipolar disorder is a chronic and severe mental health disorder. Early stratification of individuals into subgroups based on age at onset (AAO) has the potential to inform diagnosis and early intervention. Yet, the psychosocial predictors associated with AAO are unknown. AIMS: We aim to identify psychosocial factors associated with bipolar disorder AAO. METHOD: Using data from the Bipolar Disorder Research Network UK, we employed least absolute shrinkage and selection operator regression to identify psychosocial factors associated with bipolar disorder AAO. Twenty-eight factors were entered into our model, with AAO as our outcome measure. RESULTS: We included 1022 participants with bipolar disorder (µ = 23.0, s.d. ± 9.86) in our model. Six variables predicted an earlier AAO: childhood abuse (ß = -0.2855), regular cannabis use in the year before onset (ß = -0.2765), death of a close family friend or relative in the 6 months before onset (ß = -0.2435), family history of suicide (ß = -0.1385), schizotypal personality traits (ß = -0.1055) and irritable temperament (ß = -0.0685). Five predicted a later AAO: the average number of alcohol units consumed per week in the year before onset (ß = 0.1385); birth of a child in the 6 months before onset (ß = 0.2755); death of parent, partner, child or sibling in the 6 months before onset (ß = 0.3125); seeking work without success for 1 month or more in the 6 months before onset (ß = 0.3505) and a major financial crisis in the 6 months before onset (ß = 0.4575). CONCLUSIONS: The identified predictor variables have the potential to help stratify high-risk individuals into likely AAO groups, to inform treatment provision and early intervention.

The temporal dynamics of sleep disturbance and psychopathology in psychosis: a digital sampling study.

BACKGROUND: Sleep disruption is a common precursor to deterioration and relapse in people living with psychotic disorders. Understanding the temporal relationship between sleep and psychopathology is important for identifying and developing interventions which target key variables that contribute to relapse. METHODS: We used a purpose-built digital platform to sample self-reported sleep and psychopathology variables over 1 year, in 36 individuals with schizophrenia. Once-daily measures of sleep duration and sleep quality, and fluctuations in psychopathology (positive and negative affect, cognition and psychotic symptoms) were captured. We examined the temporal relationship between these variables using the Differential Time-Varying Effect (DTVEM) hybrid exploratory-confirmatory model. RESULTS: Poorer sleep quality and shorter sleep duration maximally predicted deterioration in psychosis symptoms over the subsequent 1-8 and 1-12 days, respectively. These relationships were also mediated by negative affect and cognitive symptoms. Psychopathology variables also predicted sleep quality, but not sleep duration, and the effect sizes were smaller and of shorter lag duration. CONCLUSIONS: Reduced sleep duration and poorer sleep quality anticipate the exacerbation of psychotic symptoms by approximately 1-2 weeks, and negative affect and cognitive symptoms mediate this relationship. We also observed a reciprocal relationship that was of shorter duration and smaller magnitude. Sleep disturbance may play a causal role in symptom exacerbation and relapse, and represents an important and tractable target for intervention. It warrants greater attention as an early warning sign of deterioration, and low-burden, user-friendly digital tools may play a role in its early detection.

Changes in daily mental health service use and mortality at the commencement and lifting of COVID-19 'lockdown' policy in 10 UK sites: a regression discontinuity in time design.

OBJECTIVES: To investigate changes in daily mental health (MH) service use and mortality in response to the introduction and the lifting of the COVID-19 'lockdown' policy in Spring 2020. DESIGN: A regression discontinuity in time (RDiT) analysis of daily service-level activity. SETTING AND PARTICIPANTS: Mental healthcare data were extracted from 10 UK providers. OUTCOME MEASURES: Daily (weekly for one site) deaths from all causes, referrals and discharges, inpatient care (admissions, discharges, caseloads) and community services (face-to-face (f2f)/non-f2f contacts, caseloads): Adult, older adult and child/adolescent mental health; early intervention in psychosis; home treatment teams and liaison/Accident and Emergency (A&E). Data were extracted from 1 Jan 2019 to 31 May 2020 for all sites, supplemented to 31 July 2020 for four sites. Changes around the commencement and lifting of COVID-19 'lockdown' policy (23 March and 10 May, respectively) were estimated using a RDiT design with a difference-in-difference approach generating incidence rate ratios (IRRs), meta-analysed across sites. RESULTS: Pooled estimates for the lockdown transition showed increased daily deaths (IRR 2.31, 95% CI 1.86 to 2.87), reduced referrals (IRR 0.62, 95% CI 0.55 to 0.70) and reduced inpatient admissions (IRR 0.75, 95% CI 0.67 to 0.83) and caseloads (IRR 0.85, 95% CI 0.79 to 0.91) compared with the pre lockdown period. All community services saw shifts from f2f to non-f2f contacts, but varied in caseload changes. Lift of lockdown was associated with reduced deaths (IRR 0.42, 95% CI 0.27 to 0.66), increased referrals (IRR 1.36, 95% CI 1.15 to 1.60) and increased inpatient admissions (IRR 1.21, 95% CI 1.04 to 1.42) and caseloads (IRR 1.06, 95% CI 1.00 to 1.12) compared with the lockdown period. Site-wide activity, inpatient care and community services did not return to pre lockdown levels after lift of lockdown, while number of deaths did. Between-site heterogeneity most often indicated variation in size rather than direction of effect. CONCLUSIONS: MH service delivery underwent sizeable changes during the first national lockdown, with as-yet unknown and unevaluated consequences.