Masters candidate in Biostatistics, Xuan Pu, will present:

“Constructing One-Stage and Two-Stage Multi-Protein Classifiers for Early-Stage Ovarian Cancer Screening”

Plan B Adviser: Ashley Petersen

Abstract: Because of the lack of early-stage diagnosis and mild symptoms, most ovarian cancer patients are diagnosed at late stages, leading to a low survival rate. This research aimed to improve the early-stage ovarian cancer screening performance by constructing a one-stage multi-protein classifier, and possibly simplify screening procedures by exploring the use of a two-stage classifier using a dataset with 116 early-stage ovarian cancer samples and 336 healthy controls. We first constructed a multi-protein classifier using LASSO with 10-fold cross-validation, which identified the four most contributing proteins: ITGAV, SEZ6L, WFDC2, and MUC.16 (CA125). The AUC of the multi-protein classifier was 0.974 (95% CI: 0.951-0.989), which was higher than the CA125 alone classifier (AUC: 0.958, 95% CI: 0.927-0.982). The sensitivity at 95% specificity of multi-protein classifier was 0.914 (95% CI: 0.849-0.964), which was also higher than the CA125 alone classifier (0.879, 95% CI: 0.798-0.942). We then constructed a two-stage classifier that classified the study population into healthy, uncertain, or ovarian cancer groups solely based on CA125 cutoff values in stage one. Then a multi-protein classifier was constructed at stage two to make predictions for those in the uncertain group. The optimal cutoff values were chosen to be a lower cutoff value of 2.1 and an upper cutoff value of 8.4. The two-stage classifier had a higher AUC (0.988) compared to the one-stage multi-protein classifier and it especially increased the sensitivities at high specificities. Multi-protein classifiers may be a promising way to improve early-stage ovarian cancer screening and two-stage classifiers may help to simplify early-stage ovarian cancer screening by decreasing the unnecessary financial and emotional burdens of additional testing.

Refreshments will be served prior to the presentation.