If your biomarker Kaplan-Meier plot only shows two curves, you're missing the real story.
Most teams I work with think they've found a predictive biomarker when they see separation between biomarker-positive and biomarker-negative patients.
But here's the problem: that could just be prognostic (as in the illustrated example).
A prognostic biomarker tells you about patient outcomes regardless of treatment. A predictive biomarker tells you who will respond to YOUR drug specifically.
The difference? It determines whether you have a precision medicine strategy or just wishful thinking.
To truly validate predictive value, your Kaplan-Meier should include four curves:
✅ Your drug in biomarker-positive patients
✅ Your drug in biomarker-negative patients
✅ Standard of care in biomarker-positive patients
✅ Standard of care in biomarker-negative patients
Only when your test agent shows differential treatment effect between biomarker subgroups—differing from standard of care—can you claim predictive value.
In fact distinguishing prognostic from predictive effect can be quite complex, especially when a biomarker can be both. To read more about this, see the FDA Best resource:
