How Early-Career Researchers and Patients can collaborate in AI-in-healthcare research
By Duncan Reynolds
In a newly published research article in Research Involvement and Engagement, Duncan Reynolds and colleagues show how Patient and Public Involvement (PPIE) was able to target everyday decision makers on an AI-in-healthcare research project.
Artificial intelligence (AI) is already predicting cancer, transcribing GP appointments, and being used to provide physiotherapy in the NHS. Yet, many of these AI tools were built without involvement from those who they would be used to treat, the patients. Patient and Public Involvement and Engagement (PPIE) is meant to solve this gap, but too often it is limited to reviewing work after key decisions have been made, leading to tokenistic contributions whilst researchers miss out on key real-world insights.
In large AI projects, everyday choices about data cleaning, variable selection and model tuning are often not made by principal investigators but by doctoral students and post-docs – the Early-Career Researchers (ECRs). The idea became, target them, and you can influence the engine room of the research. This led to the NIHR-funded AI MULTIPLY consortium creating twice-monthly, one-hour online sessions where an ECR presents work-in-progress to up to twenty PPIE contributors. The idea behind this was that ECRs could show their work in progress, ask specific questions to the contributors about their research direction to make sure it was in line with patient benefit, and then go and make changes based on the sessions.
The paper provides ethnographic accounts, combined with survey data, to show how the sessions ran in practice and the impact they had on the project as well as the ECRs. Examples of this are Hassan (name pseudonymised) who made concrete changes to his work based on his session where he added benchmarking his model against existing risk scores, as well as survey results showing how ECRs had gained confident in their communication and presentation skills.
From this work, we recommend the following for those who wish to set up a similar initiative:
- Identify which members of the research team are making everyday decisions.
For AI MULTIPLY, this was ECRs engaging in data work, but this may not be the case in other settings. Therefore, those who are making everyday decisions need to be identified so that opportunities for engagement are not lost. In the findings, we showed how research decisions were directly impacted by this work.
- Have a clear structure in place and secure senior support.
The sessions in the AI MULTIPLY project were twice monthly and had the support of the project and work package leads. This was important during difficulties of signing up ECRs, as PIs were able and willing to use their influence to encourage engagement.
- Establish a culture where those presenting felt empowered to present work-in-progress, and where PPIE members were encouraged to share insights candidly.
It was normalised to share early stages and unfinished work (as seen in the findings). This was important to allow early involvement from PPIE. Further to this, support was provided from experienced PPIE members so that researchers could have a point of access if they have questions or concerns before and/or after their sessions. As many of the targeted researchers were new to PPIE work on AI MULTIPLY, this helped mitigate some of the apprehension around engaging with a new type of work.
Normalise sharing of unfinished code and analyses
- Allocate appropriate resources for the sessions to take place.
This includes finances to renumerate contributors for their time, and admin support to help with the setting up of meetings, answering of emails, and experienced facilitation for the meetings to support the ECRs to develop these skills. This important for all PPIE but is often overlooked.
- Be flexible based on constant feedback.
People’s lives and circumstances can change, and flexibility to allow for a range of contribution is necessary. In the case of AI MULTIPLY, sessions were originally planned on Fridays, but due to feedback from ECRs and contributors that this was inappropriate for Muslim members of the team as they would be at prayers, the sessions were moved.
The full article is available to read here.