AI important for future personalised care

“It is therefore of great value, both for the healthcare but also for society at large, to understand how and when these AI techniques can be used most effectively,” says Alexander Galozy, PhD student at the School of Information Technology at Halmstad University.

Alexander Galozy will present his research within the field at his licentiate seminar on April 19. He is part of the research group at the University that, together with Region Halland, runs the iMedA project (Improving Medication Adherence through Person Centered Care and Adaptive Intervention). The researchers believe that there is a need to develop individualised digital solutions to support patients with high blood pressure, so-called hypertension, to improve their medication adherence. Patients with high blood pressure have good refill adherence. However, collecting one's medicine is not necessarily the same as actually taking it. Therefore, the researchers hope that AI and individualised digital tools can be one way to support the patient group.

Mobile application as support

The project is developing a mobile application providing personalised digital interventions, which will consider different behaviours and needs since the reasons for not following a treatment plan differ between individuals.

The first step has been to identify various reasons why patients with high blood pressure do not follow the prescribed treatment plan. The researchers have done this by analysing large amounts of electronic health records (EHR). As a part of the iMedA project, Alexander Galozy has, in his licentiate thesis, analysed the problem with measuring medication adherence using current EHR. He also has investigated how machine learning models can be used on EHR to predict patients' medical adherence.

“Measuring medication adherence using real-world EHR is not easy. There is a lot that can complicate the work, for example, through duplicates and missing data. This makes the analyses incorrect. However, we have begun to find solutions to solve the problems,” says Alexander Galozy.

Machine learning needs correct data

Another obstacle when predicting medical adherence behaviour is the data generated from patients who visit their doctors frequently. It can be interpreted as these patients are more likely to refill their medication. Still, it is not certain that these patients actually pick up their medicine or follow their prescribed medical treatment.

“It complicates the work of developing machine learning models when we cannot fully trust the data. Methods to address the problem of skewed data are still lacking but are under development,” says Alexander Galozy.

Tailor-made reminders

The researchers in the iMedA project want to create persisting behavioural changes among patients with high blood pressure. Through so-called Reinforcement Learning (RL) techniques, patients will receive adapted and personalised information through the mobile application. Many solutions today depend on patients generating useful and accurate information and often take a long time before they function optimally. It can lead to the user, in this case a patient, losing interest in using the tool. As another part of Alexander Galozy's licentiate thesis, he has developed a setting and algorithm for providing adaptive personalised interventions.

“In my research, I have combined strategies to be able to personalise the solutions in the mobile health tool. Although enhanced learning looks promising, there is still a long way to go before methods become a natural part of our technical solutions,” says Alexander Galozy.

Alexander would like to continue his work in the academia and his career as a researcher in AI.

“The licentiate is just a small step in that direction, and there are still many interesting questions that I want to answer!”

Text: Anna-Frida Agardson
Top illustration: iStock
Photo: Private