Sensitivity and specificity smoothing method for determining optimal cutpoint of a continuous predictive variable

Date

2023-9-11

Author

İnce, Onur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

86
views

69
downloads

Cite This

The selection of an optimal cut-off value is crucial in health research for accurately classifying healthy and diseased individuals based on a continuous diagnostic marker. The widely-used receiver operating characteristic (ROC) curve evaluates marker performance by plotting sensitivity against specificity across various marker values. This thesis compares three methods for threshold estimation: empirical, index smoothing (IS), and sensitivity specificity smoothing (SSS). These methods are assessed using four common indexes (Youden index, concordance probability, the point closest-to-(0, 1) corner, and symmetry point), which depend on an objective function value considering sensitivity and specificity. The empirical method evaluates all cut-off points and selects the one with the optimal objective function value. IS fits a smooth curve to empirical objective function values and selects the corresponding marker value that optimizes the curve. The novel SSS method introduced here fits smooth curves to both sensitivity and specificity against the cut-off value axis. Objective function values are calculated from the estimated sensitivity and specificity values indicated by these curves, and the marker level that corresponds to the optimal value of this estimated objective function is considered the optimal cut-off point. This study assesses and compares the SSS method's performance against the empirical and IS methods for these indexes, along with evaluating performance of these four common indexes for optimal cut-off estimation. The SSS method outperforms empirical and IS methods in estimating optimal cut-off points. Among the four techniques, the Symmetry technique exhibits the highest performance, while the Youden index technique performs the least effectively.

Subject Keywords

Optimal Cut-off Point, Youden, Sensitivity, Specificity, Smoothing

URI

https://hdl.handle.net/11511/105557

Collections

Graduate School of Natural and Applied Sciences, Thesis