PredicTx: AI-powered chemotherapy precision dosing for colorectal cancer
Project lead
Dr Mahima Kalla
Client
PredictX
Opportunity
AI tools for optimising chemotherapy dosing show promise, but are rarely designed around real-world clinical workflows or patient needs. Without human-centred design, even technically capable tools risk poor adoption and unsafe translation into clinical practice.
Intervention
The Centre's research team supported the human-centred design of a user interface for PredicTx, an AI-powered precision dosing tool for colorectal cancer chemotherapy. The team conducted clinical workflow mapping, interviews with oncologists and patients to identify dosing decision challenges and care experiences, and design requirements gathering. This informed a co-design process with the PredicTx team and key stakeholders to redesign the user interface around feasible workflow entry points and cross-system integration needs.
Impact
The co-design process produced a taxonomy of 79 chemotherapy dosing information elements and six AI design considerations grounded in clinical evidence.
PredicTx was redesigned around feasible workflow entry points and cross-system integration needs, establishing a replicable pathway for translating precision oncology AI into routine care.
"What we found, coming in as a start-up, was that the Validitron team genuinely understood the territory, and in particular, the complexity of digital health and clinical AI. They helped us think about how we can co-design and validate our solution at PredicTx.""The range of expertise across the Validitron team was particularly valuable at an early stage of our project. One thing that stood out was how carefully they surfaced equity considerations, bringing in diverse stakeholder perspectives in a way that actually held together rather than feeling like an afterthought.""Their team is always so approachable and has been a real pleasure to work with. We would strongly recommend other start-ups to connect with them."– Professor Justin Yeung, Founder, PredicTx |  |