Receiver-operating characteristic curve analysis in diagnostic, prognostic and predictive biomarker research
- Kjetil Søreide, Department of Surgery, Stavanger University Hospital, P O Box 8100, Armauer Hansens vei 20, N-4068 Stavanger, Norway;
- Accepted 24 August 2008
- Published Online First 25 September 2008
Biomarkers serve various clinical functions including diagnosis, prediction and prognosis of disease.
Biomarker accuracy, including sensitivity and specificity, relies on the chosen discriminatory cut-off on a continuous test scale.
Receiver operating characteristic (ROC) curve analysis provides an objective statistical method to assess the diagnostic accuracy of a test with a continuous outcome by graphically displaying the trade-offs of the true-positive rate (sensitivity) and false-positive rate (1-specificity).
ROC analysis is integral to modern biomarker research, as recommended in the REMARK guidelines.
Clinical vigilance is needed when applying ROC analysis for biomarkers for a time-to-event study.
From a clinical perspective, biomarkers may have a variety of functions, which correspond to different stages (table 1) in disease development, such as in the progression in cancer or cardiovascular disease.1 2 Biomarkers can assist in the care of patients who are asymptomatic (screening biomarkers), those who are suspected to have the disease (diagnostic biomarkers) and those with overt disease (prognostic biomarkers) for whom therapy may or may not have been initiated. Biomarkers can also be used for treatment response (predictive biomarkers) or surveillance after therapy (monitoring biomarkers). Fundamental for the use of biomarkers in all situations is biomarker accuracy—the ability to correctly classify one condition and/or outcome from another (eg, healthy versus diseased).
For the clinician, diagnostic testing plays a fundamental role in clinical practice. For instance, daily surgical decision making is based on the correct classification by pathology, radiology and/or clinical chemistry reports involving tissue and/or image evaluation and interpretation of disease conditions—in many decisions the interpretation is based on results in the “grey-area” although requiring “black-and-white” answers for choice of treatment (fig 1). Further, predictive modelling to estimate expected outcomes such as mortality or adverse events based on patient risk characteristics is common in any type of clinical research. Receiver-operating characteristic (ROC) …