Search

The aims of this thesis are (1) To investigate the use of new tumour markers in the preoperative diagnosis of ovarian cancer. (2) To validate previously published models and compare their performance to subjective assessment and to the models developed in this thesis. (3) To investigate the differences between small asymptomatic masses and large masses and to investigate the accuracy of published models on the diagnosis of malignancy in small masses. CA 125, CA 15-3 and CA 72-4 were significantly raised in the presence of ovarian cancer. CA 72-4 was higher in mucinous cancers and CA 125 and CA 15-3 were higher in serous and endometrioid cancers. Her-2/neu and CA 19-9 were not significantly different in benign or malignant disease. Logistic regression analysis showed age, CA125 and CA 15-3 to be the most valuable discriminators. A neural network was designed and trained which gave a sensitivity of 100% and a specificity of 90.9% on the test set. None of the six published models tested prospectively performed as well as in their original publication. The IOTA logistic regression model performed best and gave a sensitivity of 81.8% and a specificity of 72.3%. Subjective assessment of the mass gave a sensitivity of 72.7% with a specificity of 81.8%. Small masses were more commonly unilocular and large masses multilocular. Ascites, papillary proliferations, detectable flow and the smoothness of the internal wall discriminated well between benign and malignant small cysts. Age, menopausal status and CA125 were not discriminatory. None of the published models were as accurate as subjective assessment at diagnosing malignancy. These data suggest that statistical models may be of less value than tumour markers and subjective assessment in the diagnosis of ovarian malignancy.