Calculation of power in correlated receiver operating characteristic (ROC) curves:a pilot study
The purpose of this study is to evaluate power when comparing ROC curves with multi-reader and multimodality case (i.e. correlated data). We used Obuchowski’s algorithm for sample size calculation. We refered to the journal article (Oda, 2009). We computed the power increasing the AUC difference from 0.001 to 0.035 and obtained the power for the range of 0.05 to 0.927. At the point of the difference of 0.035 in the article, power was 0.865. We could see that the power increased from 0.384 to 0.865 at the point of the article when the number of readers was increased. We found that the power was enough in the article and has been changed when the number of readers was increased. We fixed the correlation between readers and modalities, thus, further investigation is required for the correlation parameter increase.
Correlated receiver operating characteristic Power Sample size Multi-reader Multi-case
Naoki Nishimoto Yoichi M. Ito Takumi Tanikawa
Center for Translational Research, Hokkaido University, Sapporo, Japan Department of Biostatistics, Hokkaido University Graduate School of Medicine, Sapporo, Japan Department of Medical Informatics, Asahikawa Medical University Hospital, Asahikawa, Japan
国际会议
The 7th Asia Pacific Association for Medical Informatics Conference(第七届亚太医药信息学大会(APAMI2012))
北京
英文
1-2
2012-10-24（万方平台首次上网日期，不代表论文的发表时间）