Evaluate Treatment Outcomes For AI-Enabled Information Collection Tool For Clinical Assessments In Mental Healthcare
Study Details
Study Description
Brief Summary
In the proposed study, the investigators aim to test an AI-prototype which adaptively collects information about a patient's mental health symptoms at the time of referral in order to support and facilitate the clinical assessment.
| Condition or Disease | Intervention/Treatment | Phase |
|---|---|---|
|
N/A |
Detailed Description
In the proposed study, the investigators aim to test an AI-prototype which adaptively collects information about a patient's mental health symptoms at the time of referral in order to support and facilitate the clinical assessment.
The AI-system consists of a machine learning model which produces a probabilistic prediction about a patient's most likely presenting problems (ranking different diagnoses based on their probability) based on standard referral information collected through Limbic Access (e.g. free-text description of the patient's symptoms, GAD-7 & PHQ-9 etc). Based on the ML prediction, up to two additional anxiety disorder specific measures (ADSM) will be administered in order to collect additional insights about the specific mental health symptoms experienced by the patient (i.e. tailored to the specific patient). The collected ADSM scores will be attached to the final referral information in order to support and facilitate the clinical assessment and ultimately improve the diagnosis process while saving clinical time. For this trial, the AI-model will only function as a support tool for the clinical assessment by collecting additional data ahead of time.
Specifically, the investigators are interested in evaluating whether the AI supported information collection improves treatment outcomes, reliability of clinical assessment, reduces waiting and assessment times as well as reduces treatment drop out rates.
Study Design
Arms and Interventions
| Arm | Intervention/Treatment |
|---|---|
| Active Comparator: Standard Limbic Access In this arm, participants will refer through the standard pathway of Limbic Access. During this process patients provide the minimal required information (e.g. demographic information) as well as some basic information about their experienced mental health symptoms (e.g. PHQ-9 & GAD-7). This information is attached to the referral provided to the clinician before the clinical assessment. |
Diagnostic Test: Standard Limbic Access pathway
Relevant information for clinical referral (e.g. demographics) and basic clinical information (e.g. PHQ-9 & Gad-7 scores) are collected during the self-referral process which is then attached to the referral notes in order to facilitate the clinical assessment conducted by the clinician.
|
| Experimental: Limbic Access with AI In this arm, provide all information as in the standard Limbic Access pathway. Based on this information a machine-learning model is used to predict the most likely presenting problem, based on which up to two additional anxiety specific measures are administered in order to collect more tailored information about the patients' experienced mental health symptoms. All the information is attached to the referral provided to the clinician before the clinical assessment. |
Diagnostic Test: Limbic Access with AI pathway
The same information as in the Limbic Access pathway is collected. However, additional information (i.e. disorder specific questionnaires) are collected for the most likely problem descriptors based on the ML-model predictions.
All information is attached to the referral in order to facilitate the clinical assessment conducted by the clinician.
|
Outcome Measures
Primary Outcome Measures
- Change from baseline depression score to after treatment [The definition of reliable and clinically significant improvement is based on a comparison of pre-treatment (at time of referral, on the day of consenting) and post-treatment (assessed at point of discharge, an average of 5 months) clinical score.]
The primary outcome will be defined as reliable and clinically significant improvement in clinical scores after treatment. Hereby, the investigators will test for changes in depression scores using Patient Health Questionnaire-9 (PHQ-9: posttreatment scores <10 and improved by ≥6 points). PHQ-9 includes 9 questions scored between 0 and 3, with higher scores indicating more severe depression.
- Change from baseline anxiety score to after treatment [The definition of reliable and clinically significant improvement is based on a comparison of pre-treatment (at time of referral, on the day of consenting) and post-treatment (assessed at point of discharge, an average of 5 months) clinical score.]
The primary outcome will be defined as reliable and clinically significant improvement in clinical scores after treatment. Hereby, we will test for changes in anxiety scores using Generalised Anxiety Disorder Assessment (GAD-7: posttreatment scores <8 and improved by ≥4 points).GAD-7 includes 7 questions scored between 0 and 3, with higher scores indicating more severe anxiety.
Secondary Outcome Measures
- Change in diagnosis [The agreement score will be based on a comparison of diagnosis at the initial assessment (before first treatment session) and the diagnoses at the end of treatment (assessed at point of discharge, an average of 5 months from referral).]
Improved diagnosis will be measured as the correspondence between the diagnosis at the initial clinic assessment and the diagnosis at the end of treatment. During treatment in IAPT the diagnoses will be continuously assessed during the course of treatment in order to step the treatment up or down if needed. The agreement of diagnoses at these two time points will be coded as a binary variable ("agreement" versus "disagreement"). The investigators will measure the percentage of patients for which the diagnosis at clinical assessment corresponds to the diagnoses at the end of treatment as a measure for the reliability for the initial diagnosis
- Clinical assessment times [This measure will be available after the clinical assessment (up to average of 1 month from consenting).]
Improved clinical efficiency will be indicated by reduced assessment times, measured by the average time per clinical assessment (in minutes).
- Waiting times [This measure will be available after the clinical assessment (up to average of 1 month from consenting).]
Patient waiting times for treatment will be measured as the time between the date of (self-referral) and the date of the clinical assessment.
- Referral Dropout Rates [During chatbot interaction (day 1)]
Patient referral dropout will be measured as any individual who consented to participate in the study, but did not complete all requested clinical information during the referral process.
- Assessment Dropout Rates [At time point of treatment termination using standard IAPT definitions (assessed up to 3 months)]
Clinical assessment dropout will be measured as any cancellation or "Did Not Attend" event for patients who successfully had a clinical assessment slot (eg. time and date) organised. The treatment cohort (Limbic Access with AI pathway) will be evaluated against a cohort of patients going through limbic Access' standard pathway across the same services and over the same time window as the study will be used for comparison.
- Treatment Dropout Rates [At time point of treatment termination using standard IAPT definitions (assessed up to 3 months)]
Treatment dropout will be measured using a "dropout" label which is added to a patient's file in the service's patient management system by the treating clinician when a dropout event occurs. The treatment cohort (Limbic Access +AI pathway) will be evaluated against a cohort of patients going through limbic Access' standard pathway across the same services and over the same time window as the study will be used for comparison.
Other Outcome Measures
- Agreement rate between the probabilistic model prediction (in the Limbic Access +AI pathway) and the clinical diagnosis. [The diagnosis of the clinician will be assessed at time of the clinical assessment (assessed up to 1 month).]
Kappa for each diagnosis will be calculated as agreement score between the model prediction and the diagnosis at clinical assessment.
- Bias in the predictive power of the model with regards to particular patient demographics [Demographic data is captured at the point of referral on the day that participants gives their consent.]
Percentage of agreement between model prediction and clinical diagnosis for different demographic groups
Eligibility Criteria
Criteria
Inclusion Criteria:
-
Participant meets minimum age requirements for the service
-
Participant's registered GP is within the IAPT CCG catchment area
Exclusion Criteria:
- Participants who are in crisis (defined by requiring urgent care or being at an urgent risk of harm)
Contacts and Locations
Locations
No locations specified.Sponsors and Collaborators
- Limbic Limited
- Insight Healthcare
Investigators
None specified.Study Documents (Full-Text)
None provided.More Information
Publications
None provided.- Limbic-303303