Leiden University PhD candidate uses AI to search patient forums for unknown side effects
Patients on patient forums share all sorts of things about their illness: from side effects and advice to messages of support. PhD candidate Anne Dirkson uses artificial intelligence (AI) to retrieve this information. ‘Ninety per cent of the side effects weren’t officially registered.’
‘Patient forums are really just like social media: all sorts of things are shared,’ Dirkson explains. ‘With a rare disease there is little chance you’ll be able to discuss it with your neighbour. And these patients often don’t know any fellow patients through others either. So a forum is useful for them.’
On the one hand, these patients share very factual information, for example how often they go for check-ups or what medication they take, and on the other, they discuss the emotional side of the disease. How do they deal with the obstacles they face in daily life, for instance in their family? Dirkson mainly focuses on the side effects mentioned by patients.
These side effects are interesting because 90 per cent of them are not reported. Dirkson researched the value of these side effects being shared. ‘Patients may share side effects that were not yet known. For them it is good that these side effects are recognised. It can also be that a certain side effect that was already known is mentioned way more often than expected. This could mean it has a big impact.’
Artificial intelligence
Dirkson used AI to optimise the filtering of side effects from patient forums. Dirkson: ‘I instructed a computer, so to speak, to learn how to recognise a side effect on a forum. If you give a computer a number of examples, it gets better and better at it.’ The computer has to recognise not only a side effect but also any synonyms. If someone writes, for example, ‘my head is pounding’, the computer has to categorise this as ‘headache’.
‘Every step was a huge step forward because it had never been done before. We are really proud.’
Alongside side effects, Dirkson looked at the advice the patients give one another. She used a rare form of stomach cancer as a case study. Extracting advice from a piece of text is more complicated than side effects because it is often vague and descriptive. ‘Every step my team and I took was a huge step forward because it had never been done before. We are really proud of this,’ Dirkson says.
Ginger tea and pickle juice
Ginger tea for nausea and pickle juice for muscle cramps: Dirkson came across all sorts on the forum. Knowledge about this advice is useful because it might make the disease more bearable for patients. ‘You do have to be careful though,’ Dirkson warns. ‘It may be that a certain piece of advice does help relieve the symptoms but also interferes with the medication.’ More research is therefore needed to determine when advice from a forum can be adopted. Dirkson hopes that a tool for this will be developed in the future. ‘Wouldn’t it be great if, as a complement to doctors, this would be a way for patients to help one another?’