SPARC, a transformer-based foundation model, was trained on 101,920,379 routine measurements from 224,047 patients with cancer to predict adverse events. Testing used held-out data, a prospective cohort of 23,126 and public datasets covering 26,012 patients. External multi-hospital cohorts of 562,620 and 206,840 patients were also used. The authors report state-of-the-art performance, and SPARC identified patients most likely to benefit from G-CSF. Fine-tuning on pooled data from several hospitals generalised better than fine-tuning on one. The abstract gives no accuracy figures or comparators, and shows no change in patient outcomes.
Why it is interesting: The model was validated prospectively and in external multi-hospital cohorts of 562,620 and 206,840 patients. It also flagged who might benefit from G-CSF, a named intervention, rather than only scoring risk.
Cite this item
BibTeX
Free sign-up. A digest of the new papers twice a week, grouped by domain.
Other drivers in Data, models and systems
-
Forecasts and their record
How the forecasts are made, how well they explained the past, and how well mortality can be forecast — and by whom.
4 this week -
What linked data reveals
What linked registries make it possible to know about whole populations
11 this week -
AI in medicine
Whether machine learning predicts, detects and decides better than the tools we have, and whether that has yet changed an outcome for a population.
5 this week -
Health systems and access
What the health system can provide and who gets it.
6 this week -
Value in health
What a treatment, health programme or policy is worth against what it costs, and how far the models that decide it can be trusted.
6 this week -
Revisions and new sources
What the figures are built from, what is newly available, and how much the numbers move when they are revised.
5 this week
