Pacific D
Learning Optimal Individualized Treatment Rules Among Ordinal Treatments with an Application to Recommended Intervals Between Blood Donations (307817)
Yuejia Xu, MRC Biostatistics Unit, University of Cambridge*Yuejia Xu, MRC Biostatistics Unit, University of Cambridge
Keywords: Blood donation, Ordinal treatment, Personalized medicine, Variable selection
There is a growing interest in personalized medicine as a way to provide better healthcare by tailoring treatments to individuals. This work was motivated by a UK-based blood donation trial (INTERVAL). The current practice in the UK is for male and female to donate every 12 and 16 weeks, respectively. The blood service aims to make blood donation safer and more sustainable by integrating baseline data from INTERVAL and developing personalized donation strategies. E.g. each female donor will be recommended to one of 16-, 14-, or 12-week inter-donation intervals. These 3 options can be viewed as 3 ordinal “treatments”. Most methods for learning optimal individualized treatment rules (ITR) can only handle 2 or multiple nominal treatments, but treating ordinal treatments as nominal may lead to information loss and suboptimal decisions. We propose a sequential re-estimation (SR) method that effectively uses treatment-ordering information to identify optimal ITR in ordinal settings. We further develop variable selection methods to handle cases with noise covariates that do not inform decision-making. The performance of SR is demonstrated via simulations and an application to INTERVAL.