Meaningful public involvement in inclusive medical AI research: a multi-phase case study

Issue

Artificial intelligence (AI) can improve care, but risks amplifying inequities if developed without lived-experience input. Despite growing recognition, patient and public involvement (PPI) in medical AI research often remains passive and late-stage. There is a lack of guidance on meaningful PPI that limits public influence on AI design and outcomes. We present a multi-phase PPI case study for developing a framework to realise inclusive medical AI (PREFER-IT), reporting the process, outcomes, and lessons against Liabo et al.’s (2020) five value principles – inclusivity, partnership, purposeful involvement, transparency, and different kinds of knowledge.

Description

Between Sept 24 and Jul 25, we embedded PPI across a transdisciplinary process to develop PREFER-IT (Ferreira et al., 2025). Involvement occurred in 4 steps: 1) pre-workshop PPI meeting (6 public partners (PP)); 2) deliberative agenda-setting via Pol.is (consensus-building platform) across PP and researchers (n = 33); 3) co-production during an in-person workshop, with 3 PP as experts; 4) structured post-workshop evaluation (n = 2).

Results

All 4 steps were mapped against Liabo’s value principles, demonstrating fulfilment across all dimensions. The PP involved (low-income and migrant backgrounds) reported feeling valued and on equal footing with academic experts, supported by proactive accessibility planning and deliberate power-sharing.

Lessons

Early, iterative, co-productive PPI of underrepresented groups is feasible in medical AI research when participation is resourced, tailored, and evaluated. Tools like Pol.is can offer a scalable, low-burden mechanism for inclusive agenda-setting, applicable across settings. Deliberative tools, combined with structured co-creation, can generate shared priorities and support meaningful PPI in medical AI research, avoiding reproducing knowledge barriers and tokenism. These lessons are transferable to teams developing or governing health AI across contexts.

Key messages

• Meaningful PPI of underrepresented groups in medical AI research is achievable through early, co-productive, multi-phase approaches that share power and tailor participation to individual needs.

• Deliberative and innovative tools, like Pol.is, offer a scalable, low-burden mechanism for equitable, data-informed agenda-setting in medical AI research, applicable across diverse contexts.