CHAPTER 8 190 Combining morphology and biomarkers Within our research group we developed a Bayesian network (ENDORISK) that can be used for preoperative risk stratification. This model includes preoperative variables; tumor grade, IHC biomarkers (ER, PR, L1CAM, p53), clinical biomarker (thrombocytosis), suspected lymph nodes on imaging, atypical endometrial cells in cervical cytology and cancer antigen (CA)125 level. Furthermore, the model includes postoperative variables; MI, LVSI, postoperative tumor grade and adjuvant treatment.21 The current ENDORISK model has been validated and demonstrated to properly identify patients with risk of LNM (area under the curve (AUC) 0.81), with a false-negative rate of 1.6% in those with very low risk of LNM (<5%).21 Validation in two independent cohorts already resulted in the similar AUC and false-negative rate.86, 87 Involving patients with SLN biopsy, did not affect the accuracy of ENDORISK.87 An advantage of Bayesian networks such as the ENDORISK-model is, it is a dynamic machine learning based computational model. Including the molecular subgroups is currently ongoing research, p53 mutant and wildtype were already included, so only POLE and MSI/ MMR will be added (Figure 1). It will be interesting to evaluate the impact of multiple classifiers in patients with high-grade EC (for example p53mut and POLE) with respect to their risk of LNM and outcome.88 In addition to the molecular subgroups and in line with the latest ESGO/ESTRO/ESP 2020 guideline and study of Creasman et al., preoperative MI will be included for the prediction of LNM.9, 34 Figure 1 shows a proposal for a revised ENDORISK model as a diagnostic algorithm for optimalisation of personalized primary treatment and it shows the doctors’ user interface.
RkJQdWJsaXNoZXIy MTk4NDMw