Anticipating Depressive and Manic Episodes in Bipolar Disorders Using Vocal Biomarkers (SPEECHBIPO)

ANTICIPATING DEPRESSIVE AND MANIC EPISODES IN BIPOLAR DISORDERS USING VOCAL BIOMARKERS

Bipolar disorder (BD) is a chronic, cyclical mental illness affecting over 1% of the global population. It is characterized by alternating episodes of elevated mood and energy (mania or hypomania) and episodes of decreased mood and energy (depression).

Manic episodes involve hyperactivity, decreased need for sleep, grandiosity, accelerated speech, and sometimes psychotic symptoms such as hallucinations or delusions. Depressive episodes, in contrast, are characterized by sadness, low energy, social withdrawal, sleep and appetite disturbances, and low self-esteem. Bipolar patients are at very high risk of suicide, with rates up to 20 times higher than in the general population; nearly half will attempt suicide during their lifetime, and 15-20% of these attempts are fatal.

BD is associated with a substantial decrease in quality of life, often greater than that seen in other mood or anxiety disorders. This reduction is primarily driven by depressive symptoms, including residual ones that may persist during remission periods. The frequent comorbidity with anxiety disorders further exacerbates the burden of the illness.

Recently, research has turned toward the concept of the digital phenotype to identify early markers of relapse using passive and continuous monitoring. Among potential digital biomarkers, voice has shown particular promise. Automated speech analysis, combined with machine learning algorithms, has demonstrated effectiveness in detecting psychiatric symptoms and differentiating mood states. In BD, vocal and linguistic patterns vary with mood fluctuations, suggesting that voice could serve as a sensitive indicator of relapse risk.

The main hypothesis of the present study is that automated analysis of speech and lifestyle data can help develop a predictive model capable of identifying early signs of relapse, whether manic, depressive, or mixed, or transitions to high-risk states in individuals with bipolar disorder.

Study Overview

• Status: Not yet recruiting
• Conditions: Bipolar Disorder (BD)
• Intervention / Treatment: Other: Voice interviews and questionnaires carried out via the CALLYOPE application; Device: Sleep measurements using an under-mattress sensor; Device: Smartwatch for measuring activity, sleep, and skin temperature

Detailed Description

Bipolar disorder (BD) is a chronic and cyclical illness that affects a significant portion of the population, representing more than 1% worldwide. It is characterized by alternating episodes of elevated mood and energy (mania or hypomania) and episodes of decreased mood and energy (depression). These mood episodes manifest as substantial variations in energy levels and behavior, which recur over time and have a major social and occupational impact. According to the World Health Organization, BD rank as the fourth leading cause of morbidity and mortality.

Manic episodes are marked by hyperactivity, exalted mood, insomnia, inflated self-esteem, expansive speech and behavior, and sometimes psychotic symptoms (such as delusions of persecution or hallucinations). In contrast, depressive episodes are characterized by low energy, sadness, social withdrawal, hypersomnia or insomnia, and low self-esteem, often accompanied by weight loss or gain and decreased or increased appetite. The risks associated with manic, depressive, or mixed episodes are numerous; notably, individuals with BD have a suicide rate up to 20 times higher than that of the general population. Nearly half of patients with BD will attempt suicide at least once in their lifetime, and 15-20% of these attempts are fatal.

BD are associated with a marked reduction in quality of life, often greater than that observed in other mood or anxiety disorders. This decrease in quality of life is more strongly correlated with depressive symptoms than with manic or hypomanic symptoms. Furthermore, poor quality of life is related to residual depressive symptoms that may persist during remission periods, as well as to the high comorbidity of bipolar disorder with anxiety disorders.

The annual relapse rate ranges between 40% and 61% during the first two years following the initiation of treatment. This high incidence of relapse makes stabilization particularly difficult for patients with BD, with a period of significant vulnerability following each episode. The average duration of hospitalization is 58 days, at an approximate cost of €850 per day, resulting in direct hospitalization costs related to mood disorder relapses of about €3 billion per year. According to the French Court of Auditors, for every euro of direct cost, there are two euros of indirect costs related to social benefits and the negative impact on employment. Extrapolating these figures, the cost of hospitalizations due to relapses in BD is estimated at €45 billion across Europe.

Moreover, each relapse or rehospitalization irreversibly affects the individual's cognitive functioning and contributes to social and occupational disintegration. Staging models of the illness based on neuroprogression have been developed, taking into account the number of relapses and the degree of functional impairment. However, these models are not yet implemented in clinical practice.

Preventing (hypo)manic and depressive episodes through early intervention is therefore a key priority both at the individual level and as a major public health issue.

A new line of research has emerged in mood disorders, focusing on the digital phenotype. Among new digital biomarkers of relapse, voice appears to be a promising parameter. Several studies have demonstrated the efficacy of automated speech analysis, using machine learning models, to aid in the diagnosis of psychiatric disorders. In bipolar disorder, the illness has been shown to influence patients' vocal and linguistic features.

Thus, the main hypothesis of the study is that automated speech analysis and lifestyle data can be used to develop a model capable of predicting either relapse (manic, depressive, or mixed episode) or the transition to a high-risk state in patients with bipolar disorder.

• Study Type: Interventional
• Enrollment (Estimated): 170
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Pierre-Alexis Geoffroy, Pr; Phone Number: +33140258263; Email: pa.geoffroy@ghu-paris.fr

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Adult patient
• Patient capable of providing informed consent
• Patient suffering from bipolar disorder according to DSM-5-TR (2022) criteria
• Patient recently discharged from hospitalization or in remission after a mood episode within the last 12 months, with a MADRS score ≤10 and a YMRS score ≤8, or based on the psychiatrist's subjective evaluation
• Patient treated with lithium/antipsychotics/benzodiazepines (monotherapy or combination therapy)
• Patient capable of performing speech assessments and responding to questionnaires on a smartphone
• Patient able to speak, read, and understand French
• Patient enrolled in a social security system

Exclusion Criteria:

• Patient with a cognitive disorder
• Patient suffering from a known demential disorder
• Patient receiving treatment for a known addictive disorder
• Patient with a condition affecting speech production
• Patient with a neurological disorder (stroke or neurodegenerative diseases)
• Patient under legal protection, guardianship, or curatorship
• Subjects deprived of liberty by judicial or administrative decision
• Pregnant or breastfeeding women

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Basic Science
• Allocation: N/A
• Interventional Model: Single Group Assignment
• Masking: None (Open Label)

Number of Arms

1

Arms and Interventions

Participant Group / Arm

Intervention / Treatment

Experimental: Patients with Bipolar Disorder (6-month observational follow-up); Performing voice and language tests on the Callyope application: testing the voice and language analysis algorithm. Daily recording (6 months) of all passive data via the connected watch and the under-mattress sensor and the number of steps via the Callyope application (remotely)

Other: Voice interviews and questionnaires carried out via the CALLYOPE application; Voice interviews carried out via the Callyope application: they consist of a series of tests, divided into two parts: Structured tasks (same content for each participant) and Semi-structured tasks (content varies for each participant). The simultaneous analysis of several speech tasks allows us to break down the different stages of speech production and the important factors that influence its achievement. In addition, patients will complete self-questionnaires via the application. Finally, lifestyle habits (number of steps) will be recorded via the application. These different tests will be carried out on the application at the inclusion visit (M0), then every week (+/- 3 days) until the end of study visit at 6 months (M6).; Device: Sleep measurements using an under-mattress sensor; The under-mattress sensor will allow continuous sleep recording (sleep duration, sleep onset and wake times, sleep apnea, sleep cycles, etc.) for patients over a 6-month period, from M0 to M6.; Device: Smartwatch for measuring activity, sleep, and skin temperature; The smartwatch will allow continuous recording of the patient's activity patterns, sleep, and skin temperature. It will be worn continuously from inclusion (M0) until the end of the study at 6 months (M6).

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Voice interviews recorded on Callyope application	The analysis of the interviews (acoustic and linguistic features) will be implemented in a voice model predicting a score.	Once a week, from Month 0 to Month 6 (end of the study visit)
Occurrence of relapses during the study period	Occurrence of relapse episodes during the 6-month follow-up period	From enrollment to the end of study at 6 months
Changes from Baseline in the depression score measured by the Montgomery-Asberg Depression Rating Scale	Depression severity will be assessed by the investigator with the MADRS (Montgomery-Asberg Depression Rating Scale) scale. Score range (min - max): 0-60 higher score relates to worse depression severity	Month 0, Month 6
Changes from Baseline in the bipolar disorder severity assessed by the Clinical Global Impression scale	Bipolar Disorder severity will be assessed by the investigator with CGI (Clinical Global Impression) scale. Score range (min - max): 0 - 7 for the severity subscale symptomatic remission correspond to a score ≤3 (CGI).	Month 0, Month 6
Changes from Baseline in the maniac symptoms severity at the Young Mania Rating Scale	Maniac symptoms severity will be assessed by the investigator with the YMRS (Young Mania Rating Scale). Score range (min - max): 0-60 higher score relates to worse maniac symptoms severity.	Month 0, Month 6

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Age	Age in years	Month 0
Weight	Weight in kilograms	Month 0
Sex	Sex (Male/Female)	Month 0
Plasma lithium concentration	Plasma lithium levels measured in mEq/L when available during the 6-month study period.	Up to 6 months (end of the study)
Plasma drug concentration	Plasma concentrations of ongoing psychotropic treatments measured when available during the 6-month study period	Up to 6 months (end of study)
Changes from Baseline in symptoms (depression, anxiety, functional autonomy, fatigue)	Symtpoms will be assessed by the score of the PHQ-9 (Patient Health Questionnaire - 9). Score range (min - max): 0 - 27 Higher score relates to a worse outcome	Once a week, from Month 0 to Month 6 (end of the study visit)
Changes from Baseline in anxiety symptoms	Anxiety symptoms will be assessed by the score of the GAD-7 (Generalized Anxiety Disorder) questionnaire. Score range (min - max): 0 - 21 Higher score relates to a worse outcome	Once a week, from Month 0 to Month 6 (end of the study visit)
Sleep disturbances severity (Athens Insomnia Scale total score)	Total score on the Athens Insomnia Scale (AIS) (range 0-24) ; Higher scores indicate more severe sleep disturbances	Once a week, from Month 0 to Month 6 (end of the study visit)
Medication adherence measured by Medication Adherence Report Scale total score	Adherence to medications will be assessed with the Medication Adherence Report Scale (MARS). A higher score indicates better therapeutic adherence.	Once a week, from Month 0 to Month 6 (end of the study visit)
Fatigue measured by the Multidimensional Fatigue Inventory	Total fatigue score on the Multidimensional Fatigue Inventory (MFI). Score range (min - max): 20 - 100; Higher total scores correspond with more acute levels of fatigue.	Month 0, Month 6
Quality of social relationships measured by the Social Network Index	Social Network Index (SNI). This questionnaire comprises a social diversity score (minimum = 0, maximum = 12), the total number of people in the social network, and the number of embedded social networks.	Month 0, Month 6
Assessment of Loneliness Using the UCLA 3-Item Loneliness Scale	Assessment of loneliness using the UCLA 3-Item Loneliness Scale. Scores range from 3 to 9, with higher scores indicating greater feelings of loneliness.	Month 0, Month 6
Mean number of steps	Collection of passive number of steps with the Callyope application and de wearable watch throughout the study	Continuous monitoring, from Month 0 to the end of study (Month 6)
Sleep duration	Measure of the total sleep time (hours) per night and the duration of sleep stages (hours) throughout the study thanks to the Withings Sleep Analyzer and wearable watch.	Continuous monitoring, from Month 0 to the end of study (Month 6)
Heart rate variability	Measure of the mean and continus heart rate variability (ms) thanks to the Withing watch.	Continuous monitoring, from Month 0 to the end of study (Month 6)
Blood oxygen saturation	Measure of the blood oxygen saturation (%) throughout the study thanks to th Withing watch	Continuous monitoring, from Month 0 to the end of study (Month 6)
Body temperature	Measure of the body temperature (°C) throughout the study thanks to the Withing watch.	Continuous monitoring, from Month 0 to the end of study (Month 6)
Sleep apnea	Measure of the apnea-hypopnea index during sleep throughout the study thanks to the Withings sleep analyzer.	Continuous monitoring, from Month 0 to the end of study (Month 6)

Collaborators and Investigators

• Sponsor: Centre Hospitalier St Anne
• Investigators: Principal Investigator: Pierre-Alexis Geoffroy, Pr, Paris Cité University; Centre ChronoS

Study record dates

Study Major Dates

• Study Start (Estimated): December 20, 2025
• Primary Completion (Estimated): May 1, 2027
• Study Completion (Estimated): May 1, 2027

Study Registration Dates

• First Submitted: December 1, 2025
• First Submitted That Met QC Criteria: December 17, 2025
• First Posted (Actual): December 23, 2025

Study Record Updates

• Last Update Posted (Actual): December 23, 2025
• Last Update Submitted That Met QC Criteria: December 17, 2025
• Last Verified: December 1, 2025

More Information

Terms related to this study

• Keywords: Bipolar Disorders; Digital Biomarkers; Relapse prediction; Speech Analysis
• Additional Relevant MeSH Terms: Bipolar and Related Disorders; Mental Disorders; Mood Disorders; Bipolar Disorder
• Other Study ID Numbers: D24-P020

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No