Introduction Early diagnosis of ovarian cancer (OC) is difficult due to the lack of effective biomarkers. Laboratory tests are necessarilly applied in clinical routine practice and some tests have shown diagnostic and prognostic relevance to OC.

Methods In this multicenter, retrospective study, we collected 98 laboratory tests and the age of women with or without OC admitted to three hospitals during 2012 and 2021. A risk prediction fusion framework (MCF model) that combined estimations from twenty artificial intelligence classification models was developed for OC diagnosis. It was evaluated on an internal validation set (3,007 individuals) and two external validation sets (5,641 and 2,344 individuals), respectively. The performance of MCF model was compared with the classic OC biomarker CA125 and HE-4, as well as seven competing state-of-the-arts methods.

Results Based on 52 features (51 laboratory tests and age), the MCF achieved an AUC of 0·949 (95% CI 0·948–0·950), 0·882 (0·880–0·885), and 0·884 (0·882–0·887) . Most features were significantly associated with accuracy of OC diagnosis according to univariate logistic regression. The MCF model showed higher AUC and sensitivity compared with CA125 and HE4 in identifying OC patients. The MCF also tolerated with input laboratory tests exclusive of CA125 or other tumor markers and yield acceptable prediction accuracy, and outperformed state-of-the-arts models. The MCF was wrapped as an OC prediction tool publicly available at https://github.com/xinzhen-lab/OC-prediction.

Conclusion/Implications MCF model using laboratory tests achieved satisfactory and consistent performance in OC diagnosis from three validation sets. The included laboratory tests besides CA125 and HE4 contributed to diagnosis of ovarian cancer.