%0 Journal Article
%T Prediction of final pathology depending on preoperative myometrial invasion and grade assessment in low-risk endometrial cancer patients: A Korean Gynecologic Oncology Group ancillary study.
%A Jang DH
%A Lee HG
%A Lee B
%A Kang S
%A Kim JH
%A Kim BG
%A Kim JW
%A Kim MH
%A Chen X
%A No JH
%A Lee JM
%A Kim JH
%A Watari H
%A Kim SM
%A Kim SH
%A Seong SJ
%A Jeong DH
%A Kim YH
%J PLoS One
%V 19
%N 6
%D 2024
%M 38935680
%F 3.752
%R 10.1371/journal.pone.0305360
%X <B>OBJECTIVE: </B>Fertility-sparing treatment (FST) might be considered an option for reproductive patients with low-risk endometrial cancer (EC). On the other hand, the matching rates between preoperative assessment and postoperative pathology in low-risk EC patients are not high enough. We aimed to predict the postoperative pathology depending on preoperative myometrial invasion (MI) and grade in low-risk EC patients to help extend the current criteria for FST.<BR><B>METHODS: </B>This ancillary study (KGOG 2015S) of Korean Gynecologic Oncology Group 2015, a prospective, multicenter study included patients with no MI or MI <1/2 on preoperative MRI and endometrioid adenocarcinoma and grade 1 or 2 on endometrial biopsy. Among the eligible patients, Groups 1-4 were defined with no MI and grade 1, no MI and grade 2, MI <1/2 and grade 1, and MI <1/2 and grade 2, respectively. New prediction models using machine learning were developed.<BR><B>RESULTS: </B>Among 251 eligible patients, Groups 1-4 included 106, 41, 74, and 30 patients, respectively. The new prediction models showed superior prediction values to those from conventional analysis. In the new prediction models, the best NPV, sensitivity, and AUC of preoperative each group to predict postoperative each group were as follows: 87.2%, 71.6%, and 0.732 (Group 1); 97.6%, 78.6%, and 0.656 (Group 2); 71.3%, 78.6% and 0.588 (Group 3); 91.8%, 64.9%, and 0.676% (Group 4).<BR><B>CONCLUSIONS: </B>In low-risk EC patients, the prediction of postoperative pathology was ineffective, but the new prediction models provided a better prediction.