Anxiety, Depressive Symptoms, Mental Health Issue
Conditions
Keywords
Artificial Intelligence, Mental Health, Telemedicine, Conversational Agent
Brief summary
Mental health concerns are a large burden for individuals, healthcare systems, and the economy. Over a million people are referred to UK mental health services each year, but more than half only receive one session of workbook-based support. Many others have to wait over 12 weeks for assessment and treatment. Wysa is a digital health app with over 3 million users that uses an artificial intelligence (AI) chatbot and a series of self-care exercises to provide mental health support and to help people develop strategies to manage their mental health and improve their resilience. This project aims to examine the impact of using Wysa on patients' symptoms of anxiety and depression during the referral process for standard UK mental health services. Patients will be given access to Wysa at the point of referral to the Improving Access to Psychological Therapies (IAPT) programme and can begin to explore the self-support tools, while they are on the waitlist for assessment and treatment. The investigators will gather a group of patients and members of the public to contribute to the recruitment of patients for the study, the methods we use to evaluate Wysa, and to provide insights on how best to share the results of our study with the general public. The investigators will use the standard IAPT measures of anxiety and depression to look at the effect of using Wysa patients' mental well-being. These questionnaires will be provided through the app and the results will be compared with a waitlist control group. The investigators will examine whether Wysa can identify people who are experiencing severe mental health difficulties so that they can be provided with additional support. Users' levels of engagement with Wysa will be assessed and some participants will be randomly selected to do an interview so the investigators can get a better understanding of what people liked and disliked about using the app and why. Finally, the investigators will be evaluating the cost-effectiveness of Wysa compared with usual care. The investigators expect that the study will show that Wysa helps reduce symptoms of anxiety and depression in people who are on the waiting list for IAPT. If the study shows this positive impact, this will provide evidence to support the use of Wysa to improve the accessibility of mental health support in clinical pathways. The investigators will be publishing the results of our study in academic journals as well as in more generally accessible platforms.
Interventions
DEVICEWysa AI chatbot mental health app
Wysa is a guided self-help and triaging tool delivered to patients via an app. It includes a chatbot and makes use of a wide range of clinically underpinned modules to provide mental health support.
Sponsors
Institute of Cancer Research, United Kingdom
Imperial College London
University of Warwick
University of Oxford
University of Plymouth
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
TREATMENT
Masking
NONE
Masking description
Due to the nature of the intervention, no blinding of participants will be possible in the study, as all participants will know whether or not they are using the app. Clinical treatment teams will also not be blinded as they will know through the electronic patient record if a participant has received the intervention. The evaluation team will be blinded to treatment using the randomisation algorithm.
Intervention model description
Participants will be randomly allocated to the intervention condition or a standard care waitlist control condition.
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Willing and able to provide informed consent; * Aged 18 years or older; * Ability to speak English to a secondary school standard; * Own a mobile device capable of supporting Wysa; * A valid email address; * Referred or self-referred to proceed through the standard IAPT care pathway.
Exclusion criteria
* Patients ineligible for the standard IAPT care pathway; * Patients with previous and current known major mental illness such as Schizophrenia, severe depression, any co-morbid neurological or neuro-psychiatric condition such as epilepsy; * Patients with current psychosis or a history of psychotic symptoms within the last 6 months; * Patients with suicidal ideation; * Patients scoring \> 15 points on PHQ 9; * Patients scoring \> 15 points on GAD-7; * Patients with significant cognitive disorders; * Patients with noted neurodevelopmental conditions such as autism or ADHD; * Patients previously diagnosed with a personality disorder; * Patients who been under the care of CMHT or a specialised mental health services in the last 2 years; * Patients who failed IAPT previously; * Patients with referrals for specialist presentations of pre-existing, diagnosed conditions requiring a specialised assessment beyond the standard clinical pathway; * Incapable of self-consent; * In a dependent/unequal relationship with the research or care teams or any PPI representatives.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Depression Severity | 3 months post-randomisation | Score on the PHQ-9, total scores range from 0 to 27 (higher scores indicated more severe depression) |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Crisis Identification | 3 months post-randomisation | Planned to compare number of users identified by the app for escalation of care compared to the number of patients in the control group who access A&E or out-of-hours services while waiting for treatment; this was not possible, data reported reflect risk events triggered by Wysa or the participant in-app |
| Uptake Rates | 3 months post-randomisation | Uptake rates of participants randomised into intervention group |
| Dropout Rates | 3 months post-randomisation | Dropout rates of participants who are randomised into intervention group and start using the app |
| Use of the App | 3-7 months post-randomisation | Participant engagement with the app (whether it was used at all based on app usage data) |
| Engagement | 3-7 months post randomisation | Qualitative feedback from semi-structured interviews about engagement with the app |
| Anxiety Severity | 3 months post-randomisation | Score on the GAD-7, total scores range from 0 to 21 (higher scores indicated more severe anxiety) |
| General Health State | Measured at baseline and 3 months post-randomisation | 5-level EQ-5D version (EQ-5D-5L) which measures health-related quality of life via 5 dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) and a visual scale - total survey scores were calculated and can range from 5 to 125 (higher scores indicate more severe health problems). A single index was obtained by transforming the ordinal survey responses using the 'eq5d' package in R. Range: The EQ-5D index score typically ranges from -0.594 to 1 (depending on the country-specific value set used). Interpretation: 1.000 represents full health (the best possible outcome). 0.000 represents a health state equivalent to death (neutral point). Negative values (e.g., -0.594) indicate health states perceived as worse than death. Higher index values indicate better health outcomes, while lower values represent poorer health states. https://euroqol.org/wp-content/uploads/2023/11/EQ-5D-5LUserguide-23-07.pdf |
| Type of App Usage - Exercises Completed | 3 months post-randomisation | Frequency and duration of app use |
| Type of App Usage - Sessions Completed | 3 months post-randomisation | Frequency and duration of app use |
| Type of App Usage - Messages to Chatbot | 3 months post-randomisation | Frequency and duration of app use |
| Patient Perceptions of Acceptability | 3-7 months post randomisation | Qualitative feedback from semi-structured interviews about the acceptability of the app |
Countries
United Kingdom
Participant flow
Participants by arm
| Arm | Count |
|---|---|
| Wysa AI Chatbot Mental Health App Wysa is a guided self-help and triaging tool delivered to patients via an app or widget. It uses Natural Language Processing to understand individuals' written inputs but not to generate responses. Wysa makes use of a wide range of clinically underpinned modules whilst gamification, clinical outcome measures and AI promotes engagement, improving efficacy and triage and reducing the cost of scale.
Wysa AI chatbot mental health app: Wysa is a guided self-help and triaging tool delivered to patients via an app. It includes a chatbot and makes use of a wide range of clinically underpinned modules to provide mental health support. | 30 |
| Waitlist Control The intervention will be compared against a waitlist group. This comparator was selected to compare Wysa to the current standard of care for patients waiting for standard IAPT assessment and treatment. | 16 |
| Total | 46 |
Withdrawals & dropouts
| Period | Reason | FG000 | FG001 |
|---|---|---|---|
| Overall Study | Failed to download app | 1 | 0 |
| Overall Study | Lost to Follow-up | 19 | 7 |
| Overall Study | Thought app was a scam | 1 | 0 |
| Overall Study | Withdrawal by Subject | 2 | 0 |
Baseline characteristics
| Characteristic | Wysa AI Chatbot Mental Health App | Waitlist Control | Total |
|---|---|---|---|
| Age, Continuous | 36.0 years STANDARD_DEVIATION 14.4 | 37.1 years STANDARD_DEVIATION 15.8 | 36.3 years STANDARD_DEVIATION 14.7 |
| GAD-7 | 11.3 units on a scale STANDARD_DEVIATION 4.92 | 11.9 units on a scale STANDARD_DEVIATION 4.92 | 11.5 units on a scale STANDARD_DEVIATION 4.87 |
| PHQ-9 | 11.9 units on a scale STANDARD_DEVIATION 5.07 | 11.0 units on a scale STANDARD_DEVIATION 4.38 | 11.6 units on a scale STANDARD_DEVIATION 4.81 |
| Race/Ethnicity, Customized Asian or Asian British | 3 Participants | 1 Participants | 4 Participants |
| Race/Ethnicity, Customized Black or Black British | 1 Participants | 0 Participants | 1 Participants |
| Race/Ethnicity, Customized Not known | 1 Participants | 2 Participants | 3 Participants |
| Race/Ethnicity, Customized Other ethnic groups | 4 Participants | 5 Participants | 9 Participants |
| Race/Ethnicity, Customized White | 21 Participants | 8 Participants | 29 Participants |
| Sex/Gender, Customized Declined to respond | 1 Participants | 0 Participants | 1 Participants |
| Sex/Gender, Customized Female | 18 Participants | 11 Participants | 29 Participants |
| Sex/Gender, Customized Male | 10 Participants | 4 Participants | 14 Participants |
| Sex/Gender, Customized Transgender female | 1 Participants | 1 Participants | 2 Participants |
Adverse events
| Event type | EG000 affected / at risk | EG001 affected / at risk |
|---|---|---|
| deaths Total, all-cause mortality | 0 / 30 | 0 / 16 |
| other Total, other adverse events | 5 / 30 | 0 / 16 |
| serious Total, serious adverse events | 0 / 30 | 0 / 16 |
Outcome results
Primary
Depression Severity
Score on the PHQ-9, total scores range from 0 to 27 (higher scores indicated more severe depression)
Time frame: 3 months post-randomisation
| Arm | Measure | Value (MEAN) | Dispersion |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Depression Severity | 9.28 units on a scale | Standard Deviation 6.56 |
| Waitlist Control | Depression Severity | 8.41 units on a scale | Standard Deviation 3.82 |
Secondary
Anxiety Severity
Score on the GAD-7, total scores range from 0 to 21 (higher scores indicated more severe anxiety)
Time frame: 3 months post-randomisation
| Arm | Measure | Value (MEAN) | Dispersion |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Anxiety Severity | 8.98 units on a scale | Standard Deviation 6.13 |
| Waitlist Control | Anxiety Severity | 8.19 units on a scale | Standard Deviation 3.22 |
Secondary
Crisis Identification
Planned to compare number of users identified by the app for escalation of care compared to the number of patients in the control group who access A&E or out-of-hours services while waiting for treatment; this was not possible, data reported reflect risk events triggered by Wysa or the participant in-app
Time frame: 3 months post-randomisation
Population: Waitlist control group was not analysed for Users triggered in-app escalation or App identified and flagged suicidal language because these participants did not have access to the app
| Arm | Measure | Group | Value (NUMBER) |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Crisis Identification | Users triggered in-app escalation | 4 participants |
| Wysa AI Chatbot Mental Health App | Crisis Identification | App identified and flagged suicidal language | 1 participants |
| Wysa AI Chatbot Mental Health App | Crisis Identification | Self-reports to the service or crisis service reports (iaptus record) | 0 participants |
| Waitlist Control | Crisis Identification | Self-reports to the service or crisis service reports (iaptus record) | 0 participants |
Secondary
Dropout Rates
Dropout rates of participants who are randomised into intervention group and start using the app
Time frame: 3 months post-randomisation
| Arm | Measure | Value (COUNT_OF_PARTICIPANTS) |
|---|---|---|
| Wysa AI Chatbot Mental Health App | Dropout Rates | 23 Participants |
| Waitlist Control | Dropout Rates | 7 Participants |
Secondary
Engagement
Qualitative feedback from semi-structured interviews about engagement with the app
Time frame: 3-7 months post randomisation
Population: Participants in the Wysa group who completed semi-structured interviews - the numbers reported are counts of the number of participants who expressed a particular theme generated from the thematic analysis
| Arm | Measure | Group | Value (NUMBER) |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Engagement | Minimal transfer of skills beyond app | 4 participants |
| Wysa AI Chatbot Mental Health App | Engagement | Reminders helped prompt use but app was easy to ignore | 4 participants |
| Wysa AI Chatbot Mental Health App | Engagement | Varying patterns of use for individuals | 4 participants |
| Wysa AI Chatbot Mental Health App | Engagement | Decrease in use over time | 2 participants |
Secondary
General Health State
5-level EQ-5D version (EQ-5D-5L) which measures health-related quality of life via 5 dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) and a visual scale - total survey scores were calculated and can range from 5 to 125 (higher scores indicate more severe health problems). A single index was obtained by transforming the ordinal survey responses using the 'eq5d' package in R. Range: The EQ-5D index score typically ranges from -0.594 to 1 (depending on the country-specific value set used). Interpretation: 1.000 represents full health (the best possible outcome). 0.000 represents a health state equivalent to death (neutral point). Negative values (e.g., -0.594) indicate health states perceived as worse than death. Higher index values indicate better health outcomes, while lower values represent poorer health states. https://euroqol.org/wp-content/uploads/2023/11/EQ-5D-5LUserguide-23-07.pdf
Time frame: Measured at baseline and 3 months post-randomisation
Population: Not all participants completed the follow-up assessment
| Arm | Measure | Group | Value (MEAN) | Dispersion |
|---|---|---|---|---|
| Wysa AI Chatbot Mental Health App | General Health State | Baseline | 0.681 units on a scale | Standard Deviation 0.163 |
| Wysa AI Chatbot Mental Health App | General Health State | 12 weeks | 0.688 units on a scale | Standard Deviation 0.174 |
| Waitlist Control | General Health State | Baseline | 0.607 units on a scale | Standard Deviation 0.235 |
| Waitlist Control | General Health State | 12 weeks | 0.644 units on a scale | Standard Deviation 0.192 |
Secondary
Patient Perceptions of Acceptability
Qualitative feedback from semi-structured interviews about the acceptability of the app
Time frame: 3-7 months post randomisation
Population: Participants in the Wysa group who completed semi-structured interviews - the numbers reported are counts of the number of participants who expressed a particular theme generated from the thematic analysis
| Arm | Measure | Group | Value (NUMBER) |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | App did not meet all needs; not long-term solution or replacement for therapy | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Easy to use but a bit overwhelming | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Tone affected user connection with app | 3 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Tool was exciting and a relief because of accessible support and expected benefit | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Trustworthiness shaped by source credibility and personal experience | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Limited chatbot frustrated users | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Range of resources were useful for users | 4 participants |
| Wysa AI Chatbot Mental Health App | Patient Perceptions of Acceptability | Suggestions for improvement | 3 participants |
Secondary
Type of App Usage - Exercises Completed
Frequency and duration of app use
Time frame: 3 months post-randomisation
Population: Participants assigned to Wysa intervention group; data from some participants was missing if they did not use that feature.
| Arm | Measure | Value (MEAN) | Dispersion |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Type of App Usage - Exercises Completed | 7.11 Exercises completed | Standard Deviation 7.79 |
Secondary
Type of App Usage - Messages to Chatbot
Frequency and duration of app use
Time frame: 3 months post-randomisation
Population: Participants assigned to Wysa intervention group; data from some participants was missing if they did not use that feature.
| Arm | Measure | Value (MEAN) | Dispersion |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Type of App Usage - Messages to Chatbot | 227 Messages to chatbot | Standard Deviation 319 |
Secondary
Type of App Usage - Sessions Completed
Frequency and duration of app use
Time frame: 3 months post-randomisation
Population: Participants assigned to Wysa intervention group; data from some participants was missing if they did not use that feature.
| Arm | Measure | Value (MEAN) | Dispersion |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Type of App Usage - Sessions Completed | 24.3 Sessions completed | Standard Deviation 29.8 |
Secondary
Uptake Rates
Uptake rates of participants randomised into intervention group
Time frame: 3 months post-randomisation
Population: Participants randomly assigned to intervention group
| Arm | Measure | Value (COUNT_OF_PARTICIPANTS) |
|---|---|---|
| Wysa AI Chatbot Mental Health App | Uptake Rates | 30 Participants |
| Waitlist Control | Uptake Rates | 16 Participants |
Secondary
Use of the App
Participant engagement with the app (whether it was used at all based on app usage data)
Time frame: 3-7 months post-randomisation
Population: Participants assigned to Wysa group
| Arm | Measure | Category | Value (COUNT_OF_PARTICIPANTS) |
|---|---|---|---|
| Wysa AI Chatbot Mental Health App | Use of the App | Engaged with Wysa | 18 Participants |
| Wysa AI Chatbot Mental Health App | Use of the App | Did not engage with Wysa | 12 Participants |