This working paper outlines OncoTwin's approach to building mechanistic tumour models that remain usable when clinical data arrives irregularly, the norm, not the exception, in many of the hospitals we're designing for.
Why this matters
Most published digital twin work assumes dense, regularly-sampled patient data: frequent imaging, continuous lab monitoring, reliable connectivity. That assumption doesn't hold in many oncology wards across East Africa, where a patient's next data point might arrive weeks after the last one.
What we're proposing
A calibration approach that explicitly models uncertainty growth between observations, rather than assuming a densely-sampled trajectory. The practical effect: the model tells you when it's extrapolating, not just what it's extrapolating to.
This publication is in preparation and has not yet been submitted for peer review. Findings described here are preliminary.