Using an AI memory game to detect depression and anxiety in older Brazilians
| ISRCTN | ISRCTN90727704 |
|---|---|
| DOI | https://doi.org/10.1186/ISRCTN90727704 |
| Secondary identifying numbers | TL 1 |
- Submission date
- 09/11/2023
- Registration date
- 10/11/2023
- Last edited
- 27/01/2025
- Recruitment status
- No longer recruiting
- Overall study status
- Completed
- Condition category
- Mental and Behavioural Disorders
Plain English summary of protocol
Background and study aims:
The n-Back Task is a cognitive test that can help identify working memory dysfunction in people with depression and anxiety. Thymia Limited, a company that specializes in developing AI models for depression and anxiety symptoms, uses n-Back performance as a biomarker to help diagnose these conditions. In a pilot study, Thymia aims to investigate the accuracy of their models in screening and diagnosing depression and anxiety symptoms. They also aim to establish the transferability of their AI model’s performance to the Lacos dataset, which represents an older Brazilian population. This dataset is distinct from the majority of Thymia’s data, which were used to train the original model. The Thymia dataset includes over 6,000 participants from mainly the UK, US, Spain, and Indonesia, aged 18 to 80 with a mean of 35 years old.
Who can participate?
Participant must be:
- Over 50 years of age
- Receiving ongoing health support via a private healthcare route at Lacos Saude
- Normal or corrected-to-normal vision
- Normal or corrected hearing
- Able to read, understand, and sign the Informed Consent Form
- Access to a laptop, smartphone, tablet, or other device and able to use them
- Willingness to be recorded
What does the study involve?
For this study, participants will take part in a series of short online activities, some of which will be repeated at regular intervals (twice a week) for 6 months. The activities can be completed on the thymia platform via the participant’s laptop or smart device from home, but must be completed on the same device each time and on a stable wifi connection. Activities include demographics and contexts questionnaires, proprietary games that record facial expressions and speech (not used in the present study), the n-Back activity of interest and standard clinical (screening) questionnaires: PHQ-8 (Patient Health Questionnaire), GAD-7 (Generalised Anxiety Disorder) and GDS (Geriatric Depression Scale).
What are the possible benefits and risks of participating?
Participants will be learning about longitudinal research participation and helping research on a future tool that might improve the assessment of mental health. If the answers the participants provide in the standardised validated questionnaires or the custom questionnaires that are clinical standard procedure show scores that are unusual, their clinician may request a call or an additional appointment with them, resulting in more personalised care.
Where is the study run from?
thymia Limited (UK)
When is the study starting and how long is it expected to run for?
August 2022 to December 2023
Who is funding the study?
thymia Limited (UK)
Lacos Saude (Brazil)
Who is the main contact?
Dr. Alexandra Georgescu
admin@thymia.ai
Contact information
Public, Scientific, Principal Investigator
International House, 64 Nile Street
London
N1 7SR
United Kingdom
 ORCID ID |
0000-0003-1929-5673 |
|---|---|
| Phone | +44 7533848443 |
| alexandra@thymia.ai |
Study information
| Study design | Single-centre observational study design |
|---|---|
| Primary study design | Observational |
| Secondary study design | Longitudinal study |
| Study setting(s) | Home |
| Study type | Diagnostic, Screening |
| Participant information sheet | Not available in web format, please use the contact details to request a patient information sheet. |
| Scientific title | Transferability of n-back task biomarkers in ai mental health models to old-age Brazilian population |
| Study acronym | TNB-AI |
| Study objectives | Primary: The AI model evaluation in the Brazilian population will achieve at least >70% AUC and >65% sensitivity and specificity in both depression and anxiety. Secondary: There is a significant difference in terms of key N-back parameters between 2-Back and 1-Back (see secondary outcome measures) in the Brazilian sample which is comparable to the existing thymia n-Back dataset. |
| Ethics approval(s) |
Approved 11/08/2022, Association of Research Managers and Administrators (7 Quy Court, Colliers Lane, Stow-cum-Quy, CB25 9AU, United Kingdom; +44 (0) 131 380 0066; djcarpenter247@gmail.com), ref: The Thymia Virtual Clinic - A User Engagement Study |
| Health condition(s) or problem(s) studied | Depressive and anxiety symptoms screening in an elderly Brazilian population (service users of a private health care provider). They may have multiple physical health comorbidities. |
| Intervention | For this study, participants will take part in a series of short online activities, some of which will be repeated at regular intervals (twice a week) for 6 months. The activities can be completed on the thymia platform via the participant’s laptop or smart device from home, but must be completed on the same device each time and on a stable wifi connection. The activities can be split into the following four categories: 1. Thymia research activities or short games, part of the current version of the thymia AI tool. These are proprietary speech eliciting tasks (prolonging vowel sounds, describing pictures, reading paragraphs of text, answering questions out loud) and the thymia version of the n-Back game (with a 1-Back and a 2-Back); 2. Thymia research questionnaires used to gather more context and information to further develop the thymia AI model [not part of the current clinical trial]. These include a tiredness questionnaire, the current state questionnaire. Our demographics questionnaire will be important in describing the sample and a user engagement questionnaire will be instrumental in gathering feedback. 3. Standardised, validated questionnaires (i.e. a widely used questionnaires). These are the mood questionnaire Patient Health Questionnaire PHQ-8 (a shortened version of the PHQ-9 that excludes questions on suicidality and self-harm), the anxiety questionnaire Generalised Anxiety Disorder GAD-7, the geriatric depression scale GDS. 4. Other/Custom questionnaires that are of clinical use and part of the clinical standard procedure like the medication adherence questionnaire and the pain scale, which are inspired by questions that clinicians typically ask during appointments and are used to further characterise the sample. |
| Intervention type | Other |
| Primary outcome measure | Measured using the thymia platform via the participant’s laptop or smart device twice a week for 6 months: 1. Area Under the Curve (AUC), specificity and sensitivity of the AI models (depression and anxiety). These are standard machine learning evaluation metrics that will help establish the performance and accuracy of a model trained on a set of data can predict accurately in an entirely new set of data collected in a clinical and completely new context. 2. PHQ-8 and GAD-7 to measure anxiety and depression |
| Secondary outcome measures | Measured using the thymia platform via the participant’s laptop or smart device twice a week for 6 months: 1. Average reaction time 2. Average omission errors 3. Average false positive rate (commission errors) 4. Average precision 5. Average recall from the n-Back game (1-Back and 2-Back) |
| Overall study start date | 01/08/2022 |
| Completion date | 31/12/2023 |
Eligibility
| Participant type(s) | Service user |
|---|---|
| Age group | Senior |
| Lower age limit | 50 Years |
| Upper age limit | 100 Years |
| Sex | Both |
| Target number of participants | 100 |
| Total final enrolment | 69 |
| Key inclusion criteria | 1. Over 50 years of age 2. Receiving ongoing health support via a private healthcare route at Lacos Saude 3. Normal or corrected-to-normal vision 4. Normal or corrected hearing 5. Able to read, understand, and sign the Informed Consent Form 6. Access to a laptop, smartphone, tablet, or other device and able to use them 7. Willingness to be recorded |
| Key exclusion criteria | 1. Under 50 years of age 2. Not receiving ongoing health support via a private healthcare route at Lacos Saude 3. No normal or corrected-to-normal vision 4. No normal or corrected hearing 5. Not able to read, understand, and sign the Informed Consent Form 6. No access to a laptop, smartphone, tablet, or other device and/or not able to use them 7. No willingness to be recorded |
| Date of first enrolment | 11/01/2023 |
| Date of final enrolment | 30/11/2023 |
Locations
Countries of recruitment
- Brazil
Study participating centre
Rio de Janeiro
20231-045
Brazil
Sponsor information
Industry
International House, 64 Nile Street
London
N1 7SR
England
United Kingdom
| Phone | +44 7776 181475 |
|---|---|
| admin@thymia.ai | |
| Website | https://thymia.ai/ |
Funders
Funder type
Industry
No information available
No information available
Results and Publications
| Intention to publish date | 31/01/2024 |
|---|---|
| Individual participant data (IPD) Intention to share | No |
| IPD sharing plan summary | Not expected to be made available |
| Publication and dissemination plan | The project's findings will be shared through multiple channels, categorised into two main groups: those targeting an academic audience and those aimed at the general public. Academic Audiences: The research team intends to generate one publication in a high-impact, peer-reviewed journal specialising in mental health and machine learning. They also plan to present their results at relevant conferences within these fields. While the publications are expected to become available 6-12 months after the trial concludes, conference presentations may be completed during the trial itself. General Public: To engage with the general public, the researchers will create a report tailored for this audience, which will be electronically accessible 6-12 months following the trial's completion, aligning with any scientific publications. This report will be shared with prominent clinical organisations. The report will also be disseminated through various social media and marketing channels. |
| IPD sharing plan | The datasets generated during and/or analysed during the current study are not expected to be made generally publicly available due to licensing and IP considerations. However, we are open to partnering with research institutes and individual academics including pseudonymised data sharing upon request. |
Study outputs
| Output type | Details | Date created | Date added | Peer reviewed? | Patient-facing? |
|---|---|---|---|---|---|
| Results article | 12/12/2024 | 27/01/2025 | Yes | No |
Editorial Notes
27/01/2025: Publication reference and total final enrolment added.
10/11/2023: Trial's existence confirmed by Association of Research Managers and Administrators.
