AI researchers in healthcare have been urged to engage with patients early in their research process to ensure they are asking the right questions and identifying the right uses for the technology.
In a chapter on AI in the Yearbook of Medical Informatics looking at the practical healthcare opportunities for the technology, section editors Annie Y. S. Lau and Pascal Staccini studied the 14 best papers on AI in healthcare in 2018 from an original group of 90 scientific articles.
The authors are concerned that much of the work on AI is not sufficiently focused on patient needs.
The paper says: “Without a clear understanding on why patients and consumers need AI in the first place, or how AI could support individuals with their healthcare needs, it is difficult to imagine the kinds of AI applications that would have meaningful and sustainable impact on individual daily lives.”
The report authors found no studies in 2018 looking at AI applications designed specifically for patients or consumers, nor any studies that elicited patient and consumer input on AI.
Focus on patients’ social media
They found that the most common use of AI for patients and consumers lay in secondary analysis of social media data such as online discussion forums.
It says that the three best papers shared a common methodology of using data-driven algorithms such as text mining and topic modelling. These were combined with insight-led approaches such as visualisation, qualitative analysis and manual review to uncover patient and consumer experiences of health and illness in online communities.
- Abdellaoui et al,  outlined a methodology to detect medication non-compliance behaviours amongst people on antidepressant and antipsychotic medications by modelling the way dosage variation and treatment interruption behaviours were discussed online.
- Jones et al.,  demonstrated it was possible to uncover the hidden, less obvious aspects of breast cancer management and recovery in online discussion forums, including aspects that are not easily ascertainable in patient clinics for example side effects while in remission, financial challenges experienced by cancer survivors over time.
- Park et al.,  identified subtle differences in the types of concerns expressed online by individuals experiencing different mental health conditions for example people with depression often discussed events associated with changes in mood whereas discussion topics amongst people with anxiety or post-traumatic stress disorder clustered around treatment- and medication-related issues.
Yet the authors are also concerned about the downsides of this social media focus.
They write: “While social media could be a good source to understand how individuals cope with and manage their conditions, they also present high risks due to widespread dissemination of poor-quality or incorrect information.
“As a result, researchers have proposed various data-driven approaches to analyse patients’ online behaviours and address the problems they experience online, such as detecting disclosure of personal health information on Twitter and determining which online health forum threads require moderator assistance.”
However, the chapter says these data-driven approaches, regardless of whether they are focused on clinicians, consumers, or patients, represent a narrow focus of AI.
Lack of direction
The authors add: “Currently, there is a lack of direction and evidence on how AI would actually benefit patients and consumers. Without a clear understanding on why patients and consumers need AI in the first place, how AI could support individuals with their healthcare needs, and what are the capabilities and limitations of AI, it is difficult to imagine the kinds of AI applications that would have meaningful and sustainable impact on individuals’ daily lives.”
The chapter does see some important opportunities.
“While many have already made headway in using data-driven and machine learning approaches in health, perhaps the challenge of AI for patients and consumers lies in how people will interact with the technology. For patients and consumers to truly benefit from AI, the design of the technology may need to be embedded deeply in their environment or perhaps even invisibly in their daily routine.”
“For example, with the rise of voice-only or voice-first interfaces, one could explore whether conversational agents have a role to support patients and consumers with their daily tasks . In addition, real-life decision support for patients and consumers remains an open opportunity provided the right problem, use case, and interaction mode are identified.”