Table 1 Parameters for evaluating clinical validity of a predictive biomarker

Clinical sensitivity

Sensitivity of the biomarker, is the ability of a biomarker or a change in biomarker to predict a meaningful change in a clinical endpoint. Sensitivity describes the relationship between the magnitude of change in the biomarker and the magnitude of change in the clinical endpoint. For example, a 50-unit increase in OncotypeDX recurrence score (RS-PCT/50) was associated with an estimated increase of 2.87 in hazard ratio (Tang et al., 2011 [21]) of distant recurrence (DRFI endpoint) in tamoxifen-treated patients.

Clinical specificity

Specificity of the biomarker, referred to as the ability of a biomarker or a change in biomarker to distinguish patients who are responders to an intervention from those who are non-responders in terms of changes in clinical endpoints. For example, the estimated hazard ratio for chemotherapy (no chemotherapy divided by chemotherapy) in the low OncotypeDX recurrence score (RS) group was 1.31 versus 0.26 in the high RS group (Tang et al., 2011 [21]), where the outcome is DRFI.

Probability of false positives

False positives occur when a desired change in a biomarker is not reflected by a positive change in a clinical endpoint or even worse, is associated with a negative change in a clinical endpoint. An example of a false positive is the detection of elevated levels of the functional or biochemical marker in the absence of clinical response to treatment. For example, a tumor that has expressed PD-L1 on the tumor cells, but does not respond to targeted anti-PD-L1 immunotherapy, is a false positive.

Probability of false negatives

False negatives occur when no change or a small observed change in a biomarker fails to signal a positive, meaningful change in a clinical endpoint; for instance a tumor that does not express PD-L1 but does respond to anti-PD-L1 immunotherapy is a false negative.

AUC

Area under ROC curve. AUC is used to compare different tests, i.e., an AUC value close to 1 indicates good discrimination, whereas an AUC of 0.5 provides no useful information regarding the likelihood of response.

ROC analysis

A graphical approach for showing accuracy across the entire range of biomarker concentrations. ROC, use to set cut points, is essentially a plot that captures true positive rate against false positive rate of an assay.

Cut point

The sensitivity and specificity of the assay must be demonstrated through robust ROC curves that provide support for the cut points established to identify responders vs. non-responders.

Hazard ratio

Chance of an event (e.g., disease recurrence, death) occurring in the treatment arm divided by the chance of the event occurring in the control arm, or vice versa.

Relative risk

Ratio of the probability of an event (e.g., disease recurrence, death) occurring in treated group to the probability of the event occurring in the control group.