Effective management of chronic conditions requires periodic clinical monitoring and frequent reevaluation of treatment decisions over the course of a patient’s illness. Therefore, evidence most suitable to inform care often involves the comparison of benefits and harm between personalized treatment plans, that is, decision policies that continuously adapt treatment to the patient’s evolving clinical condition over time. Causal inference methods to evaluate such adaptive (or, dynamic) treatment strategies were originally described for observational research studies in which investigators exert direct control over the data collection protocol. In retrospective cohort studies based on electronic health records (EHRs), monitoring events result from the joint decisions of patients and health care providers. Thus, the timing and content of EHR data can be highly varied across patients, and this variability may be driven by factors that also drive health outcomes. This variability leads to challenges but also new opportunities for evaluating dynamic treatment strategies using existing causal methods in comparative effectiveness research (CER). Variable and irregular covariate monitoring during typical longitudinal care can lead to bias concerns or can complicate the generalizability of findings to populations from other clinical settings with different monitoring practices. Because monitoring is itself a health intervention that can burden patients and health care systems, the monitoring variability in EHR data can also be exploited to evaluate the joint effects of dynamic treatment and monitoring regimens to improve the scope and relevance of CER evidence.
