Early detection of pancreatic ductal adenocarcinomas with an ensemble learning model based on a panel of protein serum biomarkers

Earlier detection of pancreatic ductal adenocarcinoma (PDAC) is key to improving patient outcomes, as it is mostly detected at advanced stages which are associated with poor survival. Developing non-invasive blood tests for early detection would be an important breakthrough. The primary objective of the work presented here was to use a unique dataset, that is both large and prospectively collected, to quantify a set of 96 cancer-associated proteins and construct multi-marker models with the capacity to accurately predict PDAC years before diagnosis. The data is part of a nested case control study within UK Collaborative Trial of Ovarian Cancer Screening and is comprised of 219 samples, collected from a total of 143 post-menopausal women who were diagnosed with pancreatic cancer within 70 months after sample collection, and 248 matched non-cancer controls. We developed a stacked ensemble modelling technique to achieve robustness in predictions and, therefore, improve performance in newly collected datasets. With a pool of 10 base-learners and a Bayesian averaging meta-learner, we can predict PDAC status with an AUC of 0.91 (95% CI 0.75 - 1.0), sensitivity of 92% (95% CI 0.54 - 1.0) at 90% specificity, up to 1 year to diagnosis, and at an AUC of 0.85 (95% CI 0.74 - 0.93) up to 2 years to diagnosis (sensitivity of 61%, 95 % CI 0.17 - 0.83, at 90% specificity). These models also use clinical covariates such as hormone replacement therapy use (at randomization), oral contraceptive pill use (ever) and diabetes and outperform biomarker combinations cited in the literature.