Anxiety, Generalized, Attention Deficit Hyperactivity Disorder (ADHD), Autism Spectrum Disorder, Bipolar Disorder (BD), Depression - Major Depressive Disorder, Obsessive Compulsive Disorder (OCD), Post Traumatic Stress Disorder, Schizophrenia Spectrum &Amp; Other Psychotic Disorders
Conditions
Keywords
Artificial Intelligence, Natural Language Processing, Acoustic Analysis, Acoustic Biomarkers, Clinical Testing, Vocal Biomarkers, Machine Learning, Mental Health, Speech Analysis
Brief summary
This study investigates whether AI-driven analysis of speech can accurately predict clinical diagnoses and assess risk for various mental or behavioral health conditions, including attention-deficit/hyperactivity disorder (ADHD), autism spectrum disorder, bipolar disorder, generalized anxiety disorder, major depressive disorder, obsessive compulsive disorder (OCD), post-traumatic stress disorder (PTSD), and schizophrenia. We aim to develop tools that can support clinicians in making more accurate and efficient diagnoses.
Interventions
DIAGNOSTIC_TESTSolicue Machine Learning Models
A comprehensive machine-learning tool aimed at providing probability estimates for several compatible disorders, including Attention Deficit Hyperactivity Disorder (ADHD), Autism Spectrum Disorder (ASD), Bipolar Affective Disorder (BPAD), Generalized Anxiety Disorder (GAD), Major Depressive Disorder (MDD), Obsessive Compulsive Disorder (OCD), Post-Traumatic Stress Disorder (PTSD), and Schizophrenia Spectrum Disorders (SSD). By offering a multi-diagnostic assessment based on speech analysis, Solicue aims to assist clinicians in navigating this complexity and potentially identifying conditions that might otherwise be overlooked in initial assessments. Solicue leverages machine learning to analyze a wide range of clinically relevant speech features, including linguistic content, prosodic elements (such as pitch, rhythm, and intonation), and other paralinguistic features.
DIAGNOSTIC_TESTMercuria Machine Learning Models
Mercuria is designed to stratify the risk of bipolar disorder in individuals presenting with depressive symptoms. This is a critical clinical need, as misdiagnosis of bipolar disorder as unipolar depression is common and can lead to inappropriate treatment, potentially worsening outcomes. By analyzing speech patterns characteristic of bipolar disorder, Mercuria aims to provide an additional tool for clinicians to differentiate between these conditions more accurately, guiding appropriate treatment decisions. Mercuria leverages machine learning to analyze a wide range of clinically relevant speech features, including linguistic content, prosodic elements (such as pitch, rhythm, and intonation), and other paralinguistic features.
Sponsors
Allwell Behavioral Health Services
The Brookline Center
Psyrin Inc.
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
13 Years to 60 Years
Inclusion criteria
1. Participants aged between 16 and 60 years. 2. Individuals currently undergoing or referred for clinical assessment of mental or behavioral health conditions (including but not limited to ADHD, ASD, BPAD, GAD, MDD, OCD, PTSD, SSD) 3. Fluent in English 4. Capable of providing informed consent, or in the case of minors, having a parent or legal guardian who can provide consent on their behalf. 5. Access to a device (smartphone, tablet, or computer) with a microphone and stable internet connectivity, necessary for completing the speech tasks.
Exclusion criteria
1. Individuals experiencing acute mental health crises or severe symptoms that would preclude meaningful participation in the study, including acute intoxication. 2. Severe cognitive impairment or intellectual disability that would prevent understanding of the study procedures or completion of the speech tasks. 3. Lack of fluency in English. 4. Technical limitations: Inability to access a suitable device or internet connection for completing the speech tasks
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Speech Battery (PSY-10) audio | At initial assessment | The speech battery consists of prompt-based tasks designed to elicit speech responses from participants in the form of monologues. This includes text reading, recall, and picture description tasks. |
| Clinical diagnosis | 0 months, 3 months, 6 months | Clinician diagnosis will be recorded for each participant at first assessment, 3-month, and 6-month follow-up. Diagnoses will be made according to ICD-11 or DSM-5 criteria for the compatible disorders: ADHD, ASD, BPAD, GAD, MDD, OCD, PTSD, and SSD. Additional relevant labels such as other mental health disorders, clinical high risk (CHR) and substance use may be recorded. |
| Performance of AI models | 0 months, 3 months, 6 months | The performance of the Mercuria and Solicue AI models will be evaluated using performance metrics of accuracy, balanced accuracy, sensitivity (recall), specificity, positive predictive value (precision), negative predictive value, F1 score, AUC-ROC. Predicted labels will be compared with the ground truth clinical diagnoses obtained from the participating mental health clinics. Confidence acceptance threshold will be set. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Patient Health Questionnaire-9 (PHQ-9) | At initial assessment | The PHQ-9 is a 9-item self-reported questionnaire that assesses the severity of depressive symptoms. |
| Reported Distress | After initial assessment | To assess the safety of online speech assessment during clinical evaluation at initial intake. The safety of online speech assessment will be measured by severity of reported distress measured using the User Feedback Form (UFF). |
| Mood Disorder Questionnaire (MDQ) | At initial assessment | The MDQ is a 15-item self-report screening instrument designed to detect bipolar spectrum disorders. It consists of 13 yes/no questions about manic symptoms, followed by two questions about the co-occurrence and impact of these symptoms. |
| DSM-5 Level 1 Cross-Cutting Symptom Measure (DSM-XC) | At initial assessment | The DSM-5 Level 1 Cross-Cutting Symptom Measure is a 23-item self-report questionnaire that screens for 13 psychiatric domains, including depression, anxiety, and substance use. |
Countries
United States
Outcome results
None listed