Casey Shiring

• 
  Student Presentation: How to Deal with Missing Data: An Application to Prostate Cancer Patients and Treatment Resistance
  Casey Shiring

  Real-life data is necessary for the application and validation of mathematical models. However, if data are missing from a dataset, the validity and usefulness of said dataset is diminished. One way to remedy this problem is by using multiple imputation - an advanced statistical method to predict the value of missing data points. As an application, we use a dataset containing clinical and other data for 109 patients through the course of a study on Intermittent Androgen Suppression therapy for prostate cancer. A model of prostate cancer treatment by Everett et al. is then fitted to the data. We examine the effects of multiple imputation on the parameter fitting and on prediction of off-treatment time span of the Everett et al. model by comparing the quality of fitting and the model’s performance in those predictions using the imputed data and using the unimputed data. Finally, we explore differences in model parameters between castration-sensitive and castration-resistant prostate cancer patients. We conclude that multiple imputation for time-series datasets improves the predictive ability of the Everett et al. model, although it does so somewhat inconsistently. Furthermore, in observing differences in parameterization between castration-sensitive and castration-resistant patients, we conclude that the androgen-independent castration-resistant cell death rate differs in a statistically significant manner between these patient types.