Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood symptomatology DOI Open Access
Jessica M. Lipschitz, Shuwen Lin, Soroush Saghafian

et al.

Acta Psychiatrica Scandinavica, Journal Year: 2024, Volume and Issue: 151(3), P. 434 - 447

Published: Oct. 13, 2024

Effective treatment of bipolar disorder (BD) requires prompt response to mood episodes. Preliminary studies suggest that predictions based on passive sensor data from personal digital devices can accurately detect episodes (e.g., between routine care appointments), but date do not use methods designed for broad application. This study evaluated whether a novel, personalized machine learning approach, trained entirely Fitbit data, with limited filtering could symptomatology in BD patients. We analyzed 54 adults BD, who wore Fitbits and completed bi-weekly self-report measures 9 months. applied (ML) models aggregated over two-week observation windows occurrences depressive (hypo)manic symptomatology, which were defined as scores above established clinical cutoffs the Patient Health Questionnaire-8 (PHQ-8) Altman Self-Rating Mania Scale (ASRM) respectively. As hypothesized, among several ML algorithms, Binary Mixed Model (BiMM) forest achieved highest area under receiver operating curve (ROC-AUC) validation process. In testing set, ROC-AUC was 86.0% depression 85.2% (hypo)mania. Using optimized thresholds calculated Youden's J statistic, predictive accuracy 80.1% (sensitivity 71.2% specificity 85.6%) 89.1% (hypo)mania 80.0% 90.1%). sound performance detecting patients using Findings expand upon evidence produce accurate predictions. Additionally, best our knowledge, this represents first application BiMM prediction. Overall, results move field step toward algorithms suitable full population patients, rather than only those high compliance, access specialized devices, or willingness share invasive data.

Language: Английский

Measuring algorithmic bias to analyze the reliability of AI tools that predict depression risk using smartphone sensed-behavioral data DOI Creative Commons
Daniel A. Adler, Caitlin A. Stamatis, Jonah Meyerhoff

et al.

npj Mental Health Research, Journal Year: 2024, Volume and Issue: 3(1)

Published: April 22, 2024

Abstract AI tools intend to transform mental healthcare by providing remote estimates of depression risk using behavioral data collected sensors embedded in smartphones. While these accurately predict elevated symptoms small, homogenous populations, recent studies show that are less accurate larger, more diverse populations. In this work, we accuracy is reduced because sensed-behaviors unreliable predictors across individuals: inconsistent demographic and socioeconomic subgroups. We first identified subgroups where a developed tool underperformed measuring algorithmic bias, with were incorrectly predicted be at lower than healthier then found inconsistencies between predictive Our findings suggest researchers developing predicting health from should think critically about the generalizability tools, consider tailored solutions for targeted

Language: Английский

Citations

6
Towards a consensus roadmap for a new diagnostic framework for mental disorders DOI Creative Commons
Martien J. Kas, Steven E. Hyman, Leanne M. Williams

et al.

European Neuropsychopharmacology, Journal Year: 2024, Volume and Issue: 90, P. 16 - 27

Published: Sept. 28, 2024

Current nosology claims to separate mental disorders into distinct categories that do not overlap with each other. This nosological separation is based on underlying pathophysiology but convention-based clustering of qualitative symptoms which are typically measured subjectively. Yet, clinical heterogeneity and diagnostic in disease dimensions within across different huge. While provide the basis for general management, they describe neurobiology gives rise individual symptomatic presentations. The ability incorporate framework stratify patients accordingly will be a critical step forward development new treatments disorders. Furthermore, it also allow physicians better understanding their illness's complexities management. To realize this ambition, paradigm shift needed build an how neuropsychiatric conditions can defined more precisely using quantitative (multimodal) biological processes markers thus significantly improve treatment success. ECNP New Frontiers Meeting 2024 set out develop consensus roadmap building by discussing its rationale, outlook, consequences all stakeholders involved. would instantiate principles procedures research could continuously precision diagnostics while moving away from traditional nosology. In meeting report, speakers' summaries presentations combined address three key elements generating such roadmap, namely, application innovative technologies, biology illness, translating approaches. general, indicated crucial need biology-informed establish precise diagnosis facilitate bringing right patient at time.

Language: Английский

Citations

3
Automated Speech Analysis in Bipolar Disorder: The CALIBER Study Protocol and Preliminary Results DOI Open Access
Gerard Anmella, Michele De Prisco, Jeremiah B. Joyce

et al.

Journal of Clinical Medicine, Journal Year: 2024, Volume and Issue: 13(17), P. 4997 - 4997

Published: Aug. 23, 2024

: Bipolar disorder (BD) involves significant mood and energy shifts reflected in speech patterns. Detecting these patterns is crucial for diagnosis monitoring, currently assessed subjectively. Advances natural language processing offer opportunities to objectively analyze them.

Language: Английский

Citations

2
Editorial: Special Issue on Digital Psychiatry DOI Open Access
Louise Birkedal Glenthøj, Maria Faurholt‐Jepsen

Acta Psychiatrica Scandinavica, Journal Year: 2024, Volume and Issue: 151(3), P. 177 - 179

Published: Dec. 11, 2024

Despite a growing recognition of mental health challenges worldwide, there remains significant gap between the demand for and availability services. The WHO estimates that globally, up to 71% individuals with severe illnesses such as psychosis receive no treatment, access is even more limited in low-income countries. Barriers stigma, resource shortages, insufficiently trained professionals may exacerbate this issue [1, 2]. Given resources available, recent report by World Health Organization stated "the use mobile wireless technologies (mhealth) support achievement objectives has potential transform face service delivery across globe" [3]. On global scale, it not feasible propose practices based entirely on in-person care will ever be able meet need treatment. Thus, before emergence COVID-19 pandemic, was interest role new extend care. rapid advancement integration technology transforming delivery, accessibility, research methodologies. Digital tools, including wearable devices, telepsychiatric platforms, smartphone apps, virtual reality (VR), electronic record data are reshaping landscape clinical practice, research, patient engagement [4]. Similarly, digital phenotyping, artificial intelligence (AI), advanced machine learning methods offer deeper, real-time insights into patients' behaviors symptoms, potentially leading earlier diagnoses, prediction models, personalized treatment plans [5, 6]. AI-enabled programs can analyze contextualize provide information or automatically trigger actions without human interference, where machine-learning learn recognize patterns from data. These innovations address critical care, particularly pervasive capacity traditional systems need. Furthermore, solutions empower patients actively engage their through tools self-monitoring, psychoeducation, immersive, engaging interventions enhance therapeutic experience. term "digital phenotyping" been defined "moment-by-moment quantification individual-level phenotype situ using personal devices" [7, 8]. Although unanimous, some authors [9] divide phenotyping two subgroups, called "active data" "passive data." Active refer requires active input users generated, whereas passive data, sensor phone usage patterns, collected requiring any participation users. In case "objective" these inputs seen footprints traces arising "by-product" interactions technology. Self-monitored (active data) could fine-grained manner promote empowerment insight course illness early warning signs deterioration. refers approaches which gathered devices sensors analyzed physiological functions behavioral indicators [9, 10]. increased dramatically during last years, but attracted great when Tom Insel (leader NIMH until 2015) claimed, exploring further overcome [11]. An important aspect innovative intervention just-in-time adaptive (JITAI), holds enormous promoting change behavior. A JITAI covers an design adapts provision (e.g., type, content, timing, frequency) over time specific individual [12]. Continuous streams and/or passive) enable detection transitions relapse. By dynamics individual's internal state context real offers flexibly [13], enabling micro times most needed. VR another shows promise enhancing non-pharmacological various disorder [14]. creating highly realistic immersive environments, enables scenarios designed evoke cognitive, emotional, behavioral, responses. This safe, controlled setting confront manage facilitating aimed at improving functioning quality life. Through therapist-controlled visual auditory stimuli, allows individualized, gradual, fine-tuned exposure distressing triggers. It generally regarded safe minimal side effects, motion sickness dizziness [15]. Initially employed primarily anxiety disorders, expanded its application illnesses, schizophrenia spectrum disorders. Studies suggest VR-based therapies benefits whose symptoms resistant pharmacological treatments [16]. Additionally, other digitally hold particular appeal younger who often familiar proficient platforms. familiarity willingness receptiveness incorporating part Consequently, have only improve adherence also expanding broader target group. While advancements present promising opportunities, they underscore robust optimize practice. psychiatry stands intersection innovation necessity, bridging needs them effectively. manuscripts included Special Issue "Digital Psychiatry" span diverse topics, reflecting multidisciplinary nature psychiatry. From analyses studies integrating everyday contributions complement approaches. Some examples follow below. Concerning study Ambrosen et al. investigated automated computer vision facial expressions interviews 46 first-episode psychosis. Interestingly, found were associated negative initial antipsychotic response. Another interesting Dalal natural language processing (NLP) speech samples elucidate subtle deviations 147 participants healthy individuals, psychosis, high-risk schizophrenia. They established stages distinguishable each other. sophisticated analyses, Eder transdiagnostic model comparing decision tree classifiers, logistic regression, XGboost, vector predict weight gain ≥ 5% body first 4 weeks psychotropic drugs 103 psychiatric inpatients. underscored personalizing follow-up. interventions, large-scale conducted Alvarez-Jimenez 5.702 examined effectiveness moderated online social therapy platform (blended intervention) within Australian youth consistently several demonstrating improvements depression levels. Berkhof explored baseline factors characterize responders cognitive paranoia. higher safety age better outcomes reducing anxiety. addresses literature shedding light characteristics benefit novel intervention. meta-analysis Zeka current evidence treating wide range prominent Their findings highlight VR-interventions addressing conditions alcohol use, schizophrenia, anxiety, methodological rigor future reliability findings. technological application, we hope inspire clinicians researchers collaborate innovate ways ensure reaches full potential—empowering practitioners alike outcomes. peer review history article available https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/acps.13781. declare conflicts interest.

Language: Английский

Citations

1
Wearable data from students, teachers or subjects with alcohol use disorder help detect acute mood episodes via self-supervised learning (Preprint) DOI Creative Commons
Filippo Corponi, Bryan M. Li, Gerard Anmella

et al.

JMIR mhealth and uhealth, Journal Year: 2024, Volume and Issue: 12, P. e55094 - e55094

Published: May 24, 2024

Personal sensing, leveraging data passively and near-continuously collected with wearables from patients in their ecological environment, is a promising paradigm to monitor mood disorders (MDs), major determinant of the worldwide disease burden. However, collecting annotating wearable resource intensive. Studies this kind can thus typically afford recruit only few dozen patients. This constitutes one obstacles applying modern supervised machine learning techniques MD detection.

Language: Английский

Citations

0
Digital phenotyping in bipolar disorder: Using longitudinal Fitbit data and personalized machine learning to predict mood symptomatology DOI Open Access
Jessica M. Lipschitz, Shuwen Lin, Soroush Saghafian

et al.

Acta Psychiatrica Scandinavica, Journal Year: 2024, Volume and Issue: 151(3), P. 434 - 447

Published: Oct. 13, 2024

Effective treatment of bipolar disorder (BD) requires prompt response to mood episodes. Preliminary studies suggest that predictions based on passive sensor data from personal digital devices can accurately detect episodes (e.g., between routine care appointments), but date do not use methods designed for broad application. This study evaluated whether a novel, personalized machine learning approach, trained entirely Fitbit data, with limited filtering could symptomatology in BD patients. We analyzed 54 adults BD, who wore Fitbits and completed bi-weekly self-report measures 9 months. applied (ML) models aggregated over two-week observation windows occurrences depressive (hypo)manic symptomatology, which were defined as scores above established clinical cutoffs the Patient Health Questionnaire-8 (PHQ-8) Altman Self-Rating Mania Scale (ASRM) respectively. As hypothesized, among several ML algorithms, Binary Mixed Model (BiMM) forest achieved highest area under receiver operating curve (ROC-AUC) validation process. In testing set, ROC-AUC was 86.0% depression 85.2% (hypo)mania. Using optimized thresholds calculated Youden's J statistic, predictive accuracy 80.1% (sensitivity 71.2% specificity 85.6%) 89.1% (hypo)mania 80.0% 90.1%). sound performance detecting patients using Findings expand upon evidence produce accurate predictions. Additionally, best our knowledge, this represents first application BiMM prediction. Overall, results move field step toward algorithms suitable full population patients, rather than only those high compliance, access specialized devices, or willingness share invasive data.

Language: Английский

Citations

0