Analysis frame
Peer-reviewed research
Fixing the analysis before revealing survival outcomes tests generalization in a way that reconstructing a known answer cannot.
- Biomedical investigators designing replication studies
- Patients whose care could eventually be influenced by risk scores
- Performance and calibration in external populations
- Whether acting on the ranking improves a patient outcome
- A reproducible analysis can reduce the effort of independent re-testing.
- An impressive but uncalibrated ranking could invite premature clinical use.
The next experiment worth funding
Have an independent team freeze the score and test it on a differently recruited cohort. Report not only whether it ranks people, but whether its estimates are well calibrated where a real decision would be made.
Go to the source
Read the evidence behind this analysis. External links open in a new tab.
Nature Medicine — OpenScientist research and blinded evaluations


