Information from CTC measurements for metastatic breast cancer prognosis—we should do more than selecting an “optimal cut point”

Abstract

Circulating tumor cells (CTCs) are discussed as a new prognostic marker in metastatic breast cancer. Fifty to 80% of metastatic patients have circulating tumor cells in the blood circulation [1–3]. The most currently used approach for CTC detection is the CellSearch system which is an semiautomated device based on immunofluorescence and flow cytometry. CTCs are isolated by immunomagnetic beads coated with antibodies against the epithelial cell adhesion molecule (EpCAM) and identified by cytokeratinpositivity, positive nuclear staining and CD 45 negativity. In contrast to the RT–PCR based approaches (e.g. AdnaTest), CTCs can be quantified. Cristofanilli et al. [1] used the first 102 patients (training set) from a prospective multi-center trial to select a cutoff point for CTC count to stratify patients into two groups with good or poor clinical outcome and used the 75 subsequently enrolled patients as a validation set. Aiming for the clearest separation, thresholds of 1–10,000 CTCs were systematically investigated in the training set, with progression-free survival (PFS) as outcome of interest. The cut-off of 5 cells of 7.5 ml blood resulted in a good separation between the two groups in the training and validation set for PFS and overall survival (OS). Therefore, in clinical trials (e.g. SWOG S0500) the cut-off of 5 cells is used to discriminate patients with a poor clinical outcome from those with good clinical outcome and to modify treatment decisions. The approach used in the paper by Cristofanilli et al. [3] to determine an appropriate cut-off is frequently used in prognostic studies. Converting continuous variables into categorical variables and selecting the cut-point that corresponds to the most significant relation with prognosis (‘‘minimum p value approach’’) simplifies statistical analysis. No assumption about the relationship between the continuous variable and clinical outcome has to be made prior analysis. However, dichotomization has several severe disadvantages [4] including loss of information and power, uncertainty in defining a suitable cut-point, underestimation of the extent of variation in risk, and with only two groups it is impossible to detect any non linearity in the relation between the variable and outcome. The ‘‘minimum p-value approach’’ has the additional disadvantage of overestimating the size of the effect and resulting in much too small p-values. Instead of categorizing a continuous variable, Royston et al. [4] proposed to determine a suitable functional relationship between a continuous marker and the outcome. However, considering several cut-points (e.g. 3–5) gives a step function which is often suitable to exhibit the main information in the data. The cut-off points should be selected based on biological reasons, methods of determination and simplicity. In the current article by Botteri et al. [1] CTCs were determined in 80 metastatic breast cancer patients by the CellSearch assay. Four categories of CTC values (0 cells, 1–4 cells, 5–20 cells and more than 20 cells) were defined and then related to progression-free and overall survival. In addition, a restricted cubic spline regression model was applied to estimate the functional relationship between CTC and survival. A non-linear increase could be This is an invited commentary to article doi: 10.1007/s10549-009-0668-7