Abstract BACKGROUND: Cancer antigen 125 (CA125) is a membrane bound glycosylated mucin which has been reported to be the most promising biomarker for ovarian cancer screening. However, results from two large randomized trials comparing screening with CA125 and transvaginal ultrasound to usual care have shown no clinically significant difference in ovarian cancer mortality. A major limitation of CA125 as an ovarian cancer screening biomarker has been low specificity and variation between individuals by personal characteristics. Identifying personal characteristics that influence CA125 levels could be used to create personalized thresholds for CA125 thereby improving its performance as an ovarian cancer screening biomarker. We developed and conducted internal and external validation of two prediction models (linear and dichotomous) of circulating CA125 among postmenopausal women using 28,842 controls without ovarian cancer in four large population-based studies. METHODS: We identified controls from three prospective cohort studies, including Prostate, Lung, Colorectal, and Ovarian (PLCO, n=26,981), European Prospective Investigation into Cancer and Nutrition (EPIC, n=861), and the Nurses' Health Studies (NHS, n=164) as well as one population-based case-control study, the New England Case Control Study (NEC, n=1,000). CA125 was measured using the CA125II assay in PLCO, NHS, and NEC. Meso Scale Discovery (MSD) platform was used to measure CA125 in EPIC. The MSD assay values were recalibrated to the CA125II scale based on 534 NEC controls with both measurements. CA125 levels were log-transformed to achieve normal distribution or dichotomized by the upper limit of normal (35 U/ml). The prediction models were developed and internally validated using postmenopausal controls in PLCO, and then were externally validated using postmenopausal controls in EPIC, NHS and NEC. The prediction models were developed using stepwise linear or logistic regression with <0.15 as significance level for entry and retention considering factors which have been previously reported to be associated CA125 in postmenopausal women as candidate predictors (age, race, body-mass index (BMI), smoking status and duration, age at menarche, oral contraceptive use, party, age at menopause, time since menopause, hormone replacement therapy (HRT) use and duration, family history of ovarian or breast cancer, previous history of cancer, previous history of benign ovarian cyst, history of endometriosis). We then evaluated the performance of the model in the independent validation datasets. RESULTS: The linear CA125 prediction model included age, race, BMI, smoking status and duration, hysterectomy, parity, age at menopause, and duration of HRT use as predictors, explaining 5% of the variability of log-transformed CA125 levels. The correlation coefficient of the measured and predicted log-transformed CA125 was 0.18 in the PLCO testing dataset, and showed comparable correlations across the independent validation datasets (0.14-0.16). The dichotomous CA125 prediction model included age, race, BMI, duration of HRT use, and hysterectomy as predictors with an AUC of 0.63 in the PLCO testing dataset and 0.71 in NEC. CONCLUSION: We developed linear and dichotomous circulating CA125 prediction models in postmenopausal women that can form the foundation for creating personalized thresholds of CA125. However, other factors should be considered to increase the predictive capacity of the model. Citation Format: Naoko Sasamoto, Ana Babic, Bernard A. Rosner, Renée T. Fortner, Allison F. Vitonis, Hidemi Yamamoto, Raina N. Fichorova, Daniel W. Cramer, Rudolf Kaaks, Shelley S. Tworoger, Kathryn L. Terry. DEVELOPMENT AND VALIDATION OF CIRCULATING CA125 PREDICTION MODEL IN POSTMENOPAUSAL WOMEN WITHOUT OVARIAN CANCER [abstract]. In: Proceedings of the 12th Biennial Ovarian Cancer Research Symposium; Sep 13-15, 2018; Seattle, WA. Philadelphia (PA): AACR; Clin Cancer Res 2019;25(22 Suppl):Abstract nr DP-013.
Read full abstract