Each week, we will be publishing lay abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.
The article featured today is from Statistics in Medicine, with the full article now available to read here.
Meddis, A, Blanche, P, Bidard, FC, Latouche, A. A covariate‐specific time‐dependent receiver operating characteristic curve for correlated survival data. Statistics in Medicine. 2020; 1– 13. https://doi.org/10.1002/sim.8550
Careful validation of a prognostic biomarker is essential for the care of patients. Notably, a clinically useful prognostic biomarker allows the identification of patients at risk for a specific outcome. A first step in the evaluation of the performance of a prognostic biomarker is to identify its discriminatory ability. Discrimination quantifies how well a biomarker distinguishes patients that will experience an event from those who will be event-free within a time interval (e.g. 5 years). In addition, it is important to account for clinically relevant covariates related to the biomarker to be able to determine the added value of a new biomarker in distinguishing subjects that are considered with similar risk profiles.The covariate-specific time dependent ROC curve and its AUC are widely used statistical tools which quantify the discrimination of a candidate biomarker adjusting for some relevant covariates. However, clustered data often arises in medical research and observations collected in the same cluster tend to be correlated because of some common shared features and, in this context of clustered survival data, it is not clear, so far, how to proceed in the assessment of the discrimination for a biomarker. In this work, a new estimate of the covariate-specific time dependent ROC curve and its AUC is proposed, where a shared frailty model to take into account the clustered structure of the data is considered. Furthermore, the performance of the nonparametric method with inverse probability censoring weighting (IPCW) is evaluated showing that it is an eligible approach in presence of discrete covariate. An application on non-metastatic breast cancer is illustrated. The Circulating Tumor Cells (CTCs) count is the biomarker in examination and the objective is to determine its discrimination on the time-to-death. Among the non metastatic breast cancer, patients with inflammatory tumor have a higher number of CTCs and a poorer prognosis. For refining medical decision making, it is relevant to estimate the covariate-specific ROC curves to assess the performance of CTCs within subgroups of patients having the same tumor stage. An higher discriminatory ability in group of patients with inflammatory tumor is assessed.