Cutoff Finder: a comprehensive and straightforward Web application enabling rapid biomarker cutoff optimization

Abstract

Gene or protein expression data are usually represented by metric or at least ordinal variables. In order to translate a continuous variable into a clinical decision, it is necessary to determine a cutoff point and to stratify patients into two groups each requiring a different kind of treatment. Currently, there is no standard method or standard software for biomarker cutoff determination. Therefore, we developed Cutoff Finder, a bundle of optimization and visualization methods for cutoff determination that is accessible online. While one of the methods for cutoff optimization is based solely on the distribution of the marker under investigation, other methods optimize the correlation of the dichotomization with respect to an outcome or survival variable. We illustrate the functionality of Cutoff Finder by the analysis of the gene expression of estrogen receptor (ER) and progesterone receptor (PgR) in breast cancer tissues. This distribution of these important markers is analyzed and correlated with immunohistologically determined ER status and distant metastasis free survival. Cutoff Finder is expected to fill a relevant gap in the available biometric software repertoire and will enable faster optimization of new diagnostic biomarkers. The tool can be accessed at http://molpath.charite.de/cutoff.

Figure 1. Workflow of Cutoff Finder.

The track below the icons refers to the places where the steps of data processing are done. These are the steps of data procession: 1. Data are uploaded from a tab-separated file or imported from one of three example data sets. 2. The user selects the biomarker and optionally outcome and survival variables from the table columns. 3. The user selects the method for cutoff determination. 4. The user chooses the set of plots to be generated. 5. The optimal cutoff point is determined and analysis plots are generated using R as statistical engine. 6. Cutoff point and plots are shown at the results webpage.

Figure 2. Distribution based cutoff optimization (independent of outcome and survival data) in the GSE2034 breast cancer data.

Histograms of ER (A) and PgR (B) gene expression in 286 lymph-node negative breast cancers. A mixture model of two Gaussian distributions is fitted to each of the histograms (red lines). Vertical lines designate the optimal cutoffs derived from the mixture model.

Figure 3. Cutoff optimization by correlation with a binary variable or survival in the GSE2034 breast cancer data.

(A) For each possible cutoff, ESR1 gene expression is correlated with the immunohistologically determined ER status. The odds ratio (OR) including 95% CI is plotted in dependence of the cutoff. A vertical line designates the dichotomization showing the most significant correlation with immunohistologically determined ER status. (B) For each possible cutoff, PgR gene expression was correlated with distance metastasis free survival. The hazard ratio (HR) including 95% CI is plotted in dependence of the cutoff. A vertical line designates the dichotomization showing the most significant correlation with survival. The distribution of the gene expression values in the 286 tumors is shown as a rug plot at the bottom of the figures.

Figure 4. Plot of the differences in survival time.

The mean survival time is estimated in the samples where the PgR is highly expressed and in the samples where the PgR is lowly expressed. The difference of the mean survival times including is plotted. The distribution of PgR expression is shown as rug plot at the bottom of the figure.

Figure 5. Detailed analysis of the optimal dichotomization of the GSE2034 breast cancer data.

(A) Comparison of gene expression based and immunohistochemical determination of estrogen receptor status. The classification using ESR1 expression and the optimal cutoff taken from Fig. 2A is compared to the IHC result. (B) Kaplan Meier analysis of PgR expression using the optimal cutoff taken from Fig. 2B. Distant metastasis free survival (dmfs) was significantly longer for patients with PgR expression above the cutoff.