STATISTICAL CHARACTERISTICS OF A GOOD BIOMARKER IN ONCOLOGY

OBJECTIVE: The aim of this article is to review and illustrate the attributes that analyze the performance of a predictive model, such as discrimination, calibration and clinical utility. MATERIAL AND METHODS: To illustrate a biomarker validation process, we analyzed 216 patients recruited in the Miguel Servet University Hospital, Zaragoza, Spain. The outcome of the study was clinically significant prostate cancer (Gleason >= 7). A new biomarker was built using logistic regression model from age, prostate-specific antigen, prostate volume and digital rectal exam variables. To analyze the discrimination ability, the receiver operating characteristic curve, its area under the curve (AUC), and Youden index were estimated. In addition, the calibration was analyzed through calibration curve, intercept and slope; and the clinical utility was studied by means of decision and clinical utility curves. RESULTS: The discrimination ability was good: AUC 0.790 (0.127-0.853 95% C.I.), Youden index cutoff point 0.431 (specificity 0.811, sensitivity 0.697). The Intercept was 0 and Slope 1 showing a perfect calibration. Decision curve showed good net benefit in a threshold probability range 25%-80%. Clinical utility curve showed that for a 18% cutoff point, a minimum 4.5% of CsPCa patients are wrongly classified below the cutoff point, saving 18.5% biopsies. CONCLUSIONS: A complete validation process is necessary to analyze the performance of a biomarker in oncology, based on their discrimination ability, the concordance between predicted and actual occurrence of the outcome, and its applicability in clinical practice.