An epidemiological model for prediction of endometrial cancer risk in Europe

Anika Hüsing, Laure Dossus, Pietro Ferrari, Anne Tjønneland, Louise Hansen, Guy Fagherazzi, Laura Baglietto, Helena Schock, Jenny Chang-Claude, Heiner Boeing, Annika Steffen, Antonia Trichopoulou, Christina Bamia, Michalis Katsoulis, Vittorio Krogh, Domenico Palli, Salvatore Panico, N. Charlotte Onland-Moret, Petra H. Peeters, H. Bas Bueno-de-Mesquita, Elisabete Weiderpass, Inger T. Gram, Eva Ardanaz, Mireia Obón-Santacana, Carmen Navarro, Emilio Sánchez-Cantalejo, Nerea Etxezarreta, Naomi E. Allen, Kay Tee Khaw, Nick Wareham, Sabina Rinaldi, Isabelle Romieu, Melissa A. Merritt, Marc Gunter, Elio Riboli, Rudolf Kaaks

Endometrial cancer (EC) is the fourth most frequent cancer in women in Europe, and as its incidence is increasing, prevention strategies gain further pertinence. Risk prediction models can be a useful tool for identifying women likely to benefit from targeted prevention measures. On the basis of data from 201,811 women (mostly aged 30-65 years) including 855 incident EC cases from eight countries in the European Prospective Investigation into Cancer and Nutrition cohort, a model to predict EC was developed. A step-wise model selection process was used to select confirmed predictive epidemiologic risk factors. Piece-wise constant hazard rates in 5-year age-intervals were estimated in a cause-specific competing risks model, five-fold-cross-validation was applied for internal validation. Risk factors included in the risk prediction model were body-mass index (BMI), menopausal status, age at menarche and at menopause, oral contraceptive use, overall and by different BMI categories and overall duration of use, parity, age at first full-term pregnancy, duration of menopausal hormone therapy and smoking status (specific for pre, peri- and post-menopausal women). These variables improved the discriminating capacity to predict risk over 5 years from 71 % for a model based on age alone to 77 % (overall C statistic), and the model was well-calibrated (ratio of expected to observed cases = 0.99). Our model could be used for the identification of women at increased risk of EC in Western Europe. To achieve an EC-risk model with general validity, a large-scale cohort-consortium approach would be needed to assess and adjust for population variation.