Precision Medicine Methods for Right Censored Data

Precision medicine is the tailoring of treatment plans to individual patients based on demographic, clinical, genetic, or other available data. This has the potential to greatly improve a patient’s clinical outcomes and their quality of life. Medical studies often involve time-to-event outcomes. One major challenge for analyzing time-to-event data is that some observations do not incur an event during their window of observation. Standard precision medicine methods cannot be used with right censored data without producing biased estimates. Thus, there is a need to develop precision medicine methods adapted to the censored data setting. In this manuscript we propose a number of extensions of precision medicine methods to right censored data. Extending some precision medicine methods to right censored data would require accurate prediction of the failure time. Traditional models for censored data may suffer from model misspecification and are not primarily used for prediction. Flexible machine learning methods have been extensively adopted for prediction and have received increasing interest for censored data. We develop a failure time prediction method that uses weighted support vector regression to make full use of the available data and improves performance in the presence of high censoring. A common goal in precision medicine is to identify the optimal treatment plan for patients based on their individual characteristics. We propose two individualized treatment rule estimating methods for right censored observational data. The methods are designed for situations where there are multiple treatments available, or the treatment values lie on a continuum. We develop and explore the theoretical properties of an imputation approach that could be used to extend many existing methods to right censored data.Doctor of Philosoph