Risk Prediction vs Treatment Selection
Why a test that predicts who is at higher risk is a different thing from a test that tells you which treatment will work, with examples from heart risk, circulating tumor DNA and drug-matching tests.
Content last updated . This page has not been medically reviewed.
Two questions that sound alike
When a test result is "linked to" a disease, two quite different questions are hiding inside it.- Risk prediction (prognosis): who is more likely to have a bad outcome, whatever treatment they get?
- Treatment selection (prediction of benefit): who is more likely to benefit from one specific treatment compared with another or with no treatment?
Biomarkers that answer the first question are called prognostic; those that answer the second are called predictive. A test can answer one, both, or neither. Mixing them up is one of the most common ways a test gets oversold. Evidence labels used: established, promising but limited, insufficient evidence, marketing claim.
How each is shown in a study
A biomarker is shown to be predictive of treatment benefit when the study has biomarker status on all patients, including people treated with the drug and people who were not, ideally in a randomized study, and a formal statistical test of the treatment-by-biomarker interaction is significant. [1] A purely prognostic marker is one linked to outcome regardless of treatment, with treatment effects that do not depend on the marker, and a biomarker can be both. [1]Put simply: a risk marker needs a study that follows people over time. A treatment-selection marker needs a study that compares treatments across marker groups.
Three worked examples
| Example | What it does well | What it does not show |
|---|---|---|
| hs-CRP and heart risk | Raised hs-CRP (2 mg/L or higher) is a risk-enhancing factor in guidelines [2] | That CRP picks the treatment: JUPITER gave a statin to people selected partly by CRP and cut events, but it did not compare CRP-guided selection with other ways of choosing [3] |
| ctDNA after colon cancer surgery | Positive ctDNA after surgery predicts very poor recurrence-free survival; negative predicts low recurrence risk [4] | Whether giving chemotherapy to ctDNA-positive people helps was "not well understood" going into the trial [4] |
| A drug-matching (companion diagnostic) test | Gives information essential to the safe and effective use of a particular drug [5] | Anything about other drugs or other cancers; each approval names an exact test, drug and cancer [5] |
The ctDNA example is useful because the DYNAMIC trial tested a whole strategy: patients were randomized to ctDNA-guided or standard management, and the guided group received chemotherapy less often (15% versus 28%) without worse 2-year recurrence-free survival. [4] At 5 years recurrence-free survival was 88% and 87%. [6] That is evidence about a test-and-treat strategy in stage II colon cancer, not a general statement that ctDNA selects treatment.
Prediction accuracy is not usefulness
A risk model can improve statistically yet change little in practice. In an Iceland analysis, adding a protein risk score to clinical factors raised the C index (a measure of how well a model ranks people by risk) by 0.014. [7] Whether that would change anyone's care, and whether acting on it improves outcomes, are further questions. See predictive accuracy and external validation and polygenic risk scores.Questions to ask about any test result
- Is this test telling me my risk, or telling me which treatment to choose?
- Was it tested in a trial that compared treatments, or only in people followed over time?
- Were the people studied like me (age, sex, ancestry, health status)?
- What changes if the result is high, and is that change proven to help?
- Did the study measure outcomes that matter, or only a biomarker? See biomarkers vs clinical outcomes.
Safety note
Do not start, stop or change a medicine because of a risk score. Treatment decisions belong with your prescriber: talking with a health professional.Frequently asked questions
What is the difference between a prognostic and a predictive biomarker?
A prognostic biomarker is associated with outcome regardless of treatment; a predictive one identifies who benefits from a particular treatment, which needs a treatment-by-biomarker interaction. [1]
Does a high-risk result mean a treatment will work for me?
Not necessarily. Risk and treatment benefit are separate questions, and a marker may answer only one. [1]
Is a companion diagnostic the same as a risk test?
No. The FDA describes it as providing information essential for the safe and effective use of a corresponding therapeutic product. [5]
Can a marker be both?
Yes. A biomarker may have both predictive and prognostic implications. [1]
References
- Ballman KV. Biomarker: Predictive or Prognostic? Journal of Clinical Oncology 2015. Read 2026-10-22.
- American College of Cardiology, Prioritizing Health: hsCRP (December 2025). Read 2026-10-22.
- Ridker PM et al. Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein (JUPITER). New England Journal of Medicine 2008. Read 2026-10-22.
- Tie J et al. Circulating Tumor DNA Analysis Guiding Adjuvant Therapy in Stage II Colon Cancer (DYNAMIC trial). New England Journal of Medicine 2022. Read 2026-10-22.
- U.S. Food and Drug Administration, List of FDA-Authorized Companion Diagnostic Devices (In Vitro and Imaging Tools). Read 2026-10-22.
- Tie J et al. Circulating tumor DNA analysis guiding adjuvant therapy in stage II colon cancer: 5-year outcomes of the randomized DYNAMIC trial. Nature Medicine 2025. Read 2026-10-22.
- Evaluation of Large-Scale Proteomics for Prediction of Cardiovascular Events. JAMA 2023. Read 2026-10-22.