Table 2.

Table detailing how the study objectives combined with research questions lead to the key outputs.

	Objectives	Research questions	Key outputs
1	Investigate how SMRs are currently undertaken and what barriers those undertaking them (and the patients in receipt of them) experience.	What are the barriers and facilitators to the uptake and utilisation of an AI-augmented prescribing support system for SMRs from the perspective of primary and secondary care clinicians, pharmacists, patients, and commissioners/managers involved in SMR services? What are the features that would make such a resource acceptable and usable?	Evaluation of key challenges and opportunities around medicine optimisation in general practices
2	Seek the opinions of key stakeholders involved in the SMR process about the ways in which AI approaches can be used to improve the process and identify what their requirements are for prescriber feedback systems.1.
3	Identify potential barriers/facilitators to uptake and utilisation of AI-augmented SMRs and audit and feedback dashboards for clinicians.2.
4	Curate structured clinical data from integrated records (general practice, hospital, and social care) from a variety of NHS Integrated Care Systems covering ∼11m population, adding more structured data from Natural Language Processing (NLP) of psychiatric narratives in Merseyside.	What combinations of diseases, medications, investigations and clinical contacts are associated with the greatest degree of adverse outcomes in patients with high risk of harm (frailty, co-existent mental and physical health problems and complex multimorbidity)?	A pipeline of structured and unstructured care data into multimorbidity (AI) research.
5	Use AI approaches and statistical methods to identify patterns and clusters of conditions, medications, tests, and clinical contacts preceding adverse events across three target groups then build the patterns into biostatistical causal inference and prediction of (clustered) clinical outcomes.		An AI framework for identifying those most at risk of problematic polypharmacy and for discovering disease trajectories that should trigger high priority SMRs
6	Develop visualisation methods for longitudinal summaries of multi-provider care records overlain with risk trajectories, combined with key features from AI-learned patterns/structures and clinical guidelines.	Can patient journeys over time and across care providers be adequately visualised in the context of clinical guidelines and be enriched with causal inference methods?	Novel visualisations of patient journeys enhancing medication reviews
7	Co-design a prototype tool, through iterative review and refinement of feedback systems – participating clinicians, who undertake SMRs will participate in “think-aloud” studies of the protype tool and identify positive and negative features of the tool which will allow the iterative improvement of the prototype (co-developed with patient and public representatives).	Can a learning system be created that incorporates the needs of prescribers alongside the key high-risk trajectory indicators?	Integration of outputs 1 – 4 to produce a clinically useful learning healthcare system, co-developed by the end users and supporting the delivery of SMRs by GPs and pharmacists and be accessible to patients/carers
8	Refine the later prototypes through user-group feedback and, through two workshops, to explore further the perceived strengths and weaknesses and thus the implementability of the system.