Predictive Versus Prognostic Cancer Biomarkers Explained

A prognostic biomarker tells you where a patient is likely headed. A predictive biomarker tells you whether a particular drug will change that trajectory. Put plainly, a prognostic marker is associated with outcome no matter what you do, while a predictive marker is associated with benefit from a specific therapy. That difference is not academic wordplay. It decides whether a test result should change which treatment you reach for, and the evidence needed to earn each label is different.

Two questions that sound alike but aren't

Karla Ballman set out clear working definitions in the Journal of Clinical Oncology in 2015. A prognostic biomarker informs about a likely clinical outcome, such as recurrence, progression, or death, in the absence of therapy or with a standard therapy. A predictive biomarker is associated with response, or lack of response, to a specific treatment. The two questions they answer are genuinely different: "How is this cancer likely to behave?" versus "Will this particular drug help this cancer?"

The confusion comes from the fact that a single molecule can play both roles, or neither, depending on the data behind it. A marker linked to worse survival tells you nothing, by itself, about whether any given drug will help. Yet marketing language and even some study abstracts slide from one claim to the other as if they were interchangeable. They are not, and the appraisal habit worth building is to ask which claim a given piece of evidence actually supports.

Why the interaction test is the whole game

Here is the part that separates a real predictive biomarker from a hopeful one. To show that a marker is prognostic, you need to demonstrate that it is associated with outcome regardless of treatment. To show that a marker is predictive, you need something stronger: evidence that the treatment effect itself depends on the marker. In statistical terms, that means a significant treatment-by-biomarker interaction, ideally measured in a randomized trial where some marker-positive and some marker-negative patients received the drug and others did not.

Ballman is explicit that this design is what earns the predictive label. Without a comparison arm, you cannot separate "marker-positive patients did well" from "marker-positive patients did well because of the drug." A single-arm study showing that biomarker-high patients responded more often is compatible with the marker being purely prognostic, purely predictive, or a mix. Only the interaction, tested against a control, tells them apart. This is why so many "predictive" claims deserve a second look: the study design often cannot support the word.

A useful mental check is to watch what the marker does across treatments. If a biomarker sorts patients into better and worse outcomes even when everyone gets the same treatment, that is prognostic information. If the gap in benefit between treated and untreated patients widens or narrows depending on the marker, that is predictive information. A marker can carry both signals at once, which is why the labels describe evidence, not molecules.

PD-L1 and TMB as worked examples

Immunotherapy gives two instructive cases. PD-L1, measured by immunohistochemistry on tumor or immune cells, became widely used to help select patients for immune checkpoint inhibitors. It is generally deployed as a predictive marker, since higher expression tends to track with greater likelihood of benefit from these drugs. It is also an imperfect one. Expression is measured on a continuum, cut points vary between assays and drugs, and some patients with low or absent PD-L1 still respond while some with high expression do not. The marker shifts the odds; it does not decide the case.

Tumor mutational burden, the count of mutations per megabase of coding DNA, offers the second case and a cleaner regulatory illustration. In 2020 the FDA granted accelerated, tissue-agnostic approval to pembrolizumab for adult and pediatric patients with unresectable or metastatic TMB-high solid tumors, defined as at least 10 mutations per megabase by an approved companion diagnostic. The approval, summarized in Clinical Cancer Research, rested on a prospectively planned analysis within the single-arm KEYNOTE-158 trial, where the TMB-high subset showed an overall response rate around 29 percent. That is a meaningful signal, and it earned an accelerated approval, which is a conditional pathway that asks for confirmatory evidence rather than a final verdict.

The precise reading matters here. A response rate in a single-arm study documents that TMB-high tumors respond; the absence of a randomized comparator is exactly why the predictive-versus-prognostic line stays live in ongoing debate about TMB. The threshold of 10 mutations per megabase is also a pragmatic cut on a continuous variable, and its performance varies across tumor types and sequencing platforms.

PD-L1 and TMB are frequently discussed together, which invites the assumption that they measure the same thing. Yarchoan and colleagues showed in JCI Insight that across most cancers the two are only weakly correlated and behave as largely independent biomarkers, with a model combining them explaining response better than either alone. A separate analysis in Cancers reported that the relationship between PD-L1 and TMB is inconsistent and varies by tumor type. The practical lesson is that two predictive markers can point in different directions in the same patient, and neither is a guarantee.

How to read a biomarker claim

When you meet a biomarker in a headline or a report, a few questions do most of the work. Is the claim about outcome or about benefit from a specific drug? Was the evidence a single-arm study or a randomized comparison? Was a treatment-by-biomarker interaction actually tested and significant, or was the marker only associated with response in treated patients? Is the cutoff a natural boundary or a chosen point on a continuous scale, and does it travel across assays? These questions rarely appear in promotional summaries, which is precisely why they are worth asking.

The deeper point is that "prognostic" and "predictive" are statements about the strength and design of the evidence, not fixed properties stamped on a molecule. The same marker can be prognostic in one setting, predictive in another, and unproven in a third. Holding that distinction steady is a large part of reading oncology evidence honestly.

This article is educational and not medical advice; decisions about testing or treatment belong with a patient and their own clinicians.