OnCo
ideasIdea

Whole-patient digital twins validated in prospective randomised trials

Build a computer model of each patient's cancer and body that simulates how different treatments would go, and prove in a proper trial that choosing treatment with the model helps.

Multimodal models combining genomics, pathology, imaging, pharmacokinetics and clinical history increasingly predict outcomes, but no digital twin has been validated as a decision tool in a randomised trial. The proposal is an open framework: standardised inputs, mechanistic plus learned components, calibration on federated real-world and trial data, and a series of randomised trials in which treatment selection assisted by the twin is compared with standard multidisciplinary decision-making, with regulators engaged on the evidence standard.

Hypothesis
Twin-assisted selection improves progression-free survival or reduces toxicity in at least one common indication in a randomised trial, establishing an evidence standard for oncology decision AI.
Rationale
Retrospective accuracy has not translated into clinical benefit for most oncology AI; only prospective randomised evaluation can establish whether models change outcomes, and doing it once creates the pathway.
What would test it
Randomised trial in second-line lung or colorectal cancer of twin-assisted versus standard treatment choice; primary endpoint progression-free survival, secondary toxicity and cost.
Maturity
speculative
Who has to act
research
Cost to try
Large (over $50M)
Years to first evidence
8
Bottlenecks it attacks

Connected

6top

Pages like this

not linked directly; found by shared links

technologies

2
Digital pathology & AIPathology & radiology foundation models

ideas

1
Patient-level multimodal foundation models for treatment selection

bottlenecks

3
AI that is built but not validated or deployedPreclinical models that do not predict peopleToo many combinations to test