Artificial intelligence can predict which patients will drop out of their online treatment programme

“Just by looking at when the patient logs in and out on their computer, the AI can give a heads up if the patient is likely to abort the programme they are attending,” Ulysse Côté-Allard says.

He is a researcher at the Department of Informatics at the University of Oslo.

The AI can give a warning

This can facilitate the tasks of psychologists and psychiatrists in identifying patients who require closer follow-up or a change in their treatment plan.

“We wanted to use as little information about the patients as possible. This is both to preserve the privacy of the users, and also because the AI will be easier to deploy if it does not depend on a lot of sensitive information as defined under the GDPR,” Côté-Allard says.

The only information the researchers have used to train the AI is the time of day when the patient connects to the online treatment-programme, and when they disconnect.

“We don’t rely on information such as age, sex, what the patient writes, or any kind of biometric data such as typing pattern,” he says.

Online treatment is widely used

The online treatment programme the scientists have got the collected data from is called eMeistring (link in Norwegian).

It provides guided treatment online for patients in Norway that suffer from social anxiety disorder, depression or panic disorder. The treatment is based on cognitive behavioural therapy– the basic idea of which is that by changing the way you see and think about yourself and the world affects the way you feel and behave.

“By learning about their diagnosis through reading material within the eMeistring platform, in addition to completing various exercises and questionnaires, the patients can learn coping skills which can help reduce the symptoms experienced in their daily life,” Côté-Allard explains.

The programme is guided by a mental health professional, but it is easier and much more cost efficient than having face-to-face treatment. The cost is lower for the patients themselves, and also for the healthcare provider. The patients can do the treatment when they have time.

“There are of course issues with psychological treatments delivered online, and one of them is that the reduced interaction with the therapist makes it challenging to predict patient adherence to their therapy. This is where our research comes in and why it is important. We aim to predict, as early as possible, if someone is likely to stop the treatment so that the therapist can provide closer follow-up or alter the treatment plan when they judge it necessary. In other words, the AI could work as a type of safety net,” Côté-Allard says.

The system is ready for use

The analysts have used information from 342 patients in eMeistring to train the AI using a deep neural network.

The AI starts predicting after seven days, and reaches accuracies above 70 per cent with a confidence of 95 per cent after 20 days. This corresponds to one third of the total treatment length, according to Côté-Allard.

“Taking a minimally data-sensitive approach which only uses login and logout timestamps from participants makes predicting adherence to the treatment highly challenging. Nevertheless, we think that the performance the model achieves could already be useful as a tool for clinicians in identifying patients that are more at risk of dropping out, and could thus benefit from more targeted intervention. The model's performance could also further be improved by training it on a larger cohort of patients," Côté-Allard says.

Technology in mental health

Jim Tørresen, a professor and leader of the Robotics and Intelligent Systems research group at the University of Oslo, has been leading the AI work in the project.

He is very pleased with the promising results which he expresses can be important in future therapist support systems.

“When we show that it is possible to correctly forecast adherence to a treatment programme for three out of four patients, this can contribute to therapists making better decisions early about who to prioritise for follow-up. There is also a potential in the automated systems themselves to adapt in how they interact with each patient for a best possible effect of a treatment programme,” he says.

Reference:

Côté-Allard et al. 'Adherence Forecasting for Guided Internet-Delivered Cognitive Behavioral Therapy: A Minimally Data-Sensitive Approach', IEEE Journal of Biomedical and Health Informatics, 2022. Abstract.

This article/press release is paid for and presented by the University of Oslo

This content is created by the University of Oslo's communication staff, who use this platform to communicate science and share results from research with the public. The University of Oslo is one of more than 80 owners of ScienceNorway.no. Read more here.

See more content from the University of Oslo:

• 

  A cloud of dust prevents us from seeing the universe clearly – researchers are now going to clear it up

• 

  New method tested on mice gives new hope for immunotherapy against prostate cancer

• 

  Envisioning a radically different future makes us more prepared

• 

  Some subsidies are counterproductive

• 

  Music can repair some of the damage inflicted by a prison sentence

• 

  Researchers believe that Uncle Tom's Cabin in Norwegian reflects the evolution of the racism debate