Electronic health records (EHR) contain a wealth of information that can potentially be used for research. EHR is particularly valuable for accelerating comparative effectiveness research, which seeks to estimate the relative benefit of alternative treatments in real-world settings. However there are also many challenges and limitations to conducting research with EHR-derived data, including measurement error, systematic differences between EHR and trial populations, and differential quantity and quality of data available across patients. In this dissertation, we first investigate the performance of several methods for mitigating bias due to measurement error applied to propensity scores created from error-prone covariates and provide recommendations for which methods to use in certain scenarios. Next, we develop a method to augment a randomized clinical trial with EHR data in order to create a hybrid control arm. This new method, data-adaptive-weighting, has promising properties when compared to more standard methods as well as another more recent propensity score-based method. Finally, we investigate the scenario of informed presence bias in longitudinal data, which arises when some subjects have more information in the EHR than others based on their intensity of utilization. We extend current work and explore several scenarios in which adjusting for biomarker values obtained at both pre-scheduled and patient-initiated visits in order to determine when there is risk of bias. For each topic we applied alternative methods to real-world EHR-derived data to demonstrate practical implications of alternative approaches in realistic settings.