Innovative artificial intelligence-based system for supporting the treatment of affective disorders

A major problem is also the shortage of mental health professionals, which makes it difficult for the patient to have regular contact with a doctor – in Europe, on average, there are only 18 psychiatrists per 100 000 inhabitants, with the result that a patient can count on visits on average once every three months. Medical appointments scheduled in advance do not always take place when the patient actually needs them and the treatment provided is tailored to the patient's current condition.

The low effectiveness of treatment due to the shortage of specialists and the difficulty in monitoring the patient increase the cases of hospitalisation of patients and raise the cost of treatment. For people with depression, this ranges from EUR 3,000 to EUR 5,000 per patient, while the average annual cost of treating bipolar affective disorder is around EUR 7,000. Bipolar affective disorder cost the German health service around EUR 607 million in 2015 alone. 

The main objective of the project was to use advanced machine learning algorithms to create an innovative application that could continuously and objectively monitor the risk of change of the patient’s mental state. The solution would enable more effective detection and earlier prediction of the patient's phase changes and speed up the contact with the doctor.

Effects

Clinical trials of the app have shown that it can significantly reduce hospitalisations for patients with affective disorders. During the trials of the app, no cases of  hospitalisations were reported, while in the two years preceding the trials, the number of such cases in the same group of patients was just around 12% compared to regular 40% of hospitalisation. 

Published studies indicate that continuous and effective monitoring of patients reduces the costs associated with medical care by 30% for depression and 17% for bipolar affective disorder. Reducing the risk of  hospitalisation by detecting mood phase changes earlier also reduces the patient's costs and losses due to, among other things, loss of work or difficulties in daily functioning.

Solution

Patients were provided with a clear and easy-to-use app collecting data about the patient's behaviour. The app, together with a sports bracelet paired with it, collects key behavioural markers: speech, activity, and sleep quality parameters, which are used by AI models to evaluate the patient’s mental state.

Machine learning algorithms built and trained by Britenet on historical data, analyse the collected data so that the application can send an alert about a suspected imminent change in the patient's mental state. Such an alert is sent to the patient, the attending physician, and others authorised by the patient. 

Doctors use the web application to provide clinical information to allow for continuous training of AI models. 

Additional features of both applications improved the contact of the patient with the clinician and provided information thus improving the effectiveness of care.

The MoodMon app is the first solution to use models trained at both group and individual level. Another innovative aspect of the project we have implemented is the careful selection of parameters used to analyse the patient's condition. In collaboration with specialists we have selected the 234 most important, valuable, and predictive parameters from more than a thousand different parameters – including difficult to analyse voice parameters, such as speech speed, number and length of pauses, mel cepstral coefficients, frequency-band energy, frequency distribution, and many others. It is important to stress that the application does not analyse the content of the speech nor invades the patient's privacy in any other way.  

Company background 

Project Moodmon is an advanced telemedicine platform designed to support individuals with bipolar disorder, their doctors, and families. Moodmon provides personalised interventions and continuous monitoring, thanks to advanced machine learning algorithms offers an application that can objectively monitor changes in the patient's mental state, facilitating early detection and prediction of phase changes. Validated through extensive clinical trials. Moodmon aims to revolutionise mental health care, ultimately improving outcomes for individuals with bipolar disorder.