Predicting Survivorship Appointment Nonattendance in a Community Cancer Center: A Machine-Learning Approach

Abstract

Understanding and predicting cancer survivors’ health care utilization is critical to promote quality care. The consultative system of survivorship care uses a onetime consultative appointment to transition patients from active treatment into survivorship follow-up care. Knowledge of attributes associated with nonattendance at this essential appointment is needed. An ability to predict patients with a likelihood of nonattendance would be of value to practitioners. Unfortunately, traditional data modeling techniques may not be useful in working with large numbers of variables from electronic medical record platforms. A variety of machine-learning algorithms were used to develop a model for predicting 843 survivors’ nonattendance at a comprehensive community cancer center in the southeastern United States. A parsimonious model resulted in a k-fold classification accuracy of 67.3% and included three variables. Practitioners may be able to increase utilization of follow-up care among survivors by knowing which patient groups are more likely to be survivorship appointment nonattenders.