Constructing optimal dynamic monitoring and treatment regimes: An application to hypertension care

Hypertension is a leading cause of cardiovascular diseases and morbidity, with antihypertensive drugs and blood pressure management strategies having heterogeneous effects on patients. Previous authors exploited this heterogeneity to construct optimal dynamic treatment regimes for hypertension that input patient characteristics and output the best drug or blood pressure management strategy to prescribe. There is, however, a lack of research on optimizing monitoring schedules for these patients. It is unclear whether different monitoring patterns and drug add-on strategies could lower blood pressure differently across patients. We propose a new consistent methodology to develop optimal dynamic monitoring and add-on regimes that is doubly-robust and relies on the theory of Robins' g-methods and dynamic weighted ordinary least squares. We discuss the treatment of longitudinal missing data for that inference. The approach is evaluated in large simulation studies and applied to data from the SPRINT trial in the United States to derive a new optimal rule. This type of rule could be used by patients or physicians to personalize the timing of visit and by physicians to decide whether prescribing an antihypertensive drug is beneficial.