Abstract:
|
Uncontrolled glycated hemoglobin (HbA1c) level that is higher than 7% can increase the likelihood of adverse events. Our paper is motivated from a longitudinal observational study from University of Wisconsin, using claims and enrollment files from Medicare, linked with patient electronic health records information. The goal is to develop a risk model that can identify patients who are at high-risk of uncontrolled HbA1c level in the long term in a timely fashion. The available longitudinal biomarker data will likely improve prediction. However, repeated biomarker collection could be costly and inconvenient, and risk prediction for patients at a later time could delay the necessary medical decision. We propose a cost-effective procedure that recursively incorporates comprehensive longitudinal information into a personalized risk prediction model, taking into account the cost of delaying the decision to a follow-up time when more information is available. Our method accounts for the missingness of the longitudinal biomarker and the potential non-representativeness arising from the data. Simulation studies and data analysis showed a superior performance of the proposed methods.
|