External validation of a risk stratification tool for ovarian cancer using PLCO data
• body mass index, height, age at menopause, menopausal estrogen therapy use, first-degree family history of the disease, endometriosis, and common genetic variants.
Protective factors include:
• age at menarche, parity, breastfeeding, miscarriages or pregnancy terminations (incomplete pregnancies), age at last pregnancy, tubal ligation, combined oral contraceptive use, and depot-medroxyprogesterone acetate use.
We have incorporated the exposure history of these 15 risk/protective factors to develop a risk stratification model to calculate the lifetime risk of ovarian cancer, using data from the Ovarian Cancer Association Consortium (OCAC). The model was developed in a training set comprising 80% of the OCAC dataset and was internally validated in the test set comprising the rest 20% of the data. In the internal validation, our model showed good discrimination (area under the receiver operating characteristic curve AUC=0.66) and good calibration (Hosmer-Lemeshow goodness-of-fit p>0.05). However, there is a need to validate our model using an external cohort dataset.
In this proposed study, we will validate our existing risk stratification model using data from the Prostate, Lung, and Ovarian Cancer Screening Trial (PLCO) (Aim 1). The PLCO offers a unique opportunity to conduct model optimization due to its large sample size, the detailed information collected on lifestyle factors and reproductive history, and the longitudinal design. The primary aim of this proposal is to validate the questionnaire-derived risk factors available in the PLCO. To validate the model, we will assess model discrimination, calibration, and accuracy.
To determine the impacts of incorporating novel factors on risk stratification, we will compare our risk stratification results using the risk/protective factors (described above) with a reduced model that includes only factors used in the online CanRisk ovarian cancer risk stratification model (Aim 2). CanRisk is the only model that has been approved for use by healthcare providers in the European Economic Area; no other models have been approved for clinical use elsewhere in the world to our knowledge. The CanRisk model for ovarian cancer was developed based on five rare high-penetrance mutations, a polygenic score of 36 common genetic variants, and eight environmental risk factors (including body mass index, height, tubal ligation, parity, combined oral contraceptive use, menopausal hormone therapy use, family history of ovarian cancer, and endometriosis). We will compare our full model with a reduced model that includes factors used in the CanRisk model.
1. To externally validate our risk stratification model mentioned above using data from the PLCO. We will validate the questionnaire-derived risk factors that are available in the PLCO (i.e., body mass index, height, age at menopause, menopausal estrogen therapy use, first-degree family history of the disease, endometriosis, age at menarche, parity, breastfeeding, incomplete pregnancies, tubal ligation, combined oral contraceptive use).
2. To compare the performance of our risk stratification model with the results provided by a “reduced” model using the factors used in CanRisk to determine the impact of adding additional risk factors on our risk stratification model for ovarian cancer using data from the PLCO. The “reduce” model includes the questionnaire-derived risk factors that are used in CanRisk and are available in the PLCO (i.e, body mass index, height, tubal ligation, parity, combined oral contraceptive use, menopausal hormone therapy use, family history of ovarian cancer, and endometriosis).
Britton Trabert, MS, PhD - University of Utah School of Medicine, and Huntsman Cancer Institute.
Celeste Leigh Pearce, PhD, MPH – University of Michigan School of Public Health and Rogel Cancer Center.