Objective: Receiver operating characteristic (ROC) analysis is commonly preferred for the dichotomous classification of a continuous random variable in the process of determination of the optimum cut-off point. The optimum cut-off point can be detected using ROC curve, which is a graphic presentation of the relationship between sensitivity and specificity. In the circumstances where the optimum cut-off point cannot be determined properly using ROC curves, the importance of smoothing is emphasized. In this study, it is proposed to use as an alternative locally weighted scatterplot smoothing (LOWESS) instead of ROC curves with Kernel smoothing. Methods: In our study, determination of the accurate and clear cut-off point obtained using the curves belonging to ROC and LOWESS techniques was discussed by means of an application. Excessive fluid administration during lung resection is a risk for pulmonary injury. The significant risk factors for the presence of postoperative pulmonary complications are the infusion rate of intraoperative fluids, acute respiratory distress syndrome, acute lung injury, pneumonia, atelectasis, need for toilet bronchoscopy, prolonged air leak, and failure to expand were used in the application. Results: According to the ROC analysis, the cut-off point should have been between 5.5-6mL/kg/h, but according to the LOWESS method, it was determined to be 6.125mL/kg/h. Conclusion: For the dichotomous classification, to interpret curves and to determine cut-off points perfectly, LOWESS smoothing non parametric method is strongly recommended instead of the non-parametric ROC curve.
Objective: Receiver operating characteristic (ROC) analysis is commonly preferred for the dichotomous classification of a continuous random variable in the process of determination of the optimal cut-off point. The optimal cut-off point can be detected using ROC curve, which is a graphic presentation of the relationship between sensitivity and specificity. In the circumstances where the optimal cut-off point cannot be properly determined using ROC curves, the importance of smoothing is emphasized. In this study, it is proposed to use as an alternative locally weighted scatterplot smoothing (LOWESS) instead of ROC curves with Kernel smoothing. Methods: In our study, determination of the accurate and clear cut-off point obtained using the curves belonging to ROC and LOWESS techniques was discussed by means of an application. Excessive fluid administration during lung resection is a risk for lung injury. The significant risk factors for the presence of postoperative pulmonary complications are the infusion rate of intraoperative fluids, acute respiratory distress syndrome, acute lung injury, pneumonia, atelectasis, need for toilet bronchoscopy, prolonged air leak, and failure to expand were used in the application. Results: According to the ROC analysis, the cut-off point should have been between 5.5-6mL/kg/h, but according to the LOWESS method, it was determined to be 6.125mL/kg/h. Conclusion: For the dichotomous classification, to interpret curves and to determine cut-off points perfectly, LOWESS smoothing non parametric method is strongly recommended instead of the non-parametric ROC curve.
Dergi Türü : Uluslararası
Benzer Makaleler | Yazar | # |
---|
Makale | Yazar | # |
---|