Abstract
PURPOSE: The preoperative diagnosis of endometriosis associated ovarian cancer (EAOC) remains challenging for lack of effective diagnostic biomarker. We aimed to study clinical characteristics and develop a nomogram for diagnosing EAOC before surgery. METHODS: A total of 87 patients with EAOC and 348 patients with ovarian endometrioma (OEM) were enrolled in our study. Least absolute shrinkage and selection operator (LASSO) regression and Logistic regression were utilized to select variables and construct the prediction model. The performance of the model was assessed using receiver operating characteristic (ROC) analyses and calibration plots, while decision curve analyses (DCAs) were conducted to assess clinical value. Bootstrap resampling was used to evaluated the stability of the model in the derivation set. RESULTS: The EAOC patients were older compared to the OEM patients (46.41 ± 9.62 vs. 36.49 ± 8.09 year, P < 0.001) and proportion of postmenopausal women was higher in EAOC group than in the OEM group (34.5 vs. 1.5%, P < 0.001). Our prediction model, which included age at diagnosis, tumor size, cancer antigen (CA) 19-9 and risk of ovarian malignancy algorithm (ROMA), demonstrated an area under the curve (AUC) of 0.858 (95% confidence interval (CI): 0.795-0.920) in the derivation set (N = 304) and an AUC of 0.870 (95% CI: 0.779-0.961) in the validation set (N = 131). The model fitted both the derivation (Hosmer-Lemeshow test (HL) chi-square = 12.600, P = 0.247) and the validation (HL chi-square = 8.210, P = 0.608) sets well. CONCLUSION: Compared to patients with OEM, those with EAOC exhibited distinct clinical characteristics. Our four-variable prediction model demonstrated excellent performance in both the derivation and validation sets, suggesting its potential to assist with preoperative diagnosis of EAOC.