Clinical analyses frequently draw from data used in randomized controlled trials (RCTs) and, increasingly, electronic health records (EHR). Both RCTs and EHR data can be prone to having missing observations, measurement error, or unmeasured confounding which can complicate estimation of effects of intervention or other associations present in the data. This dissertation centers on developing methods for RCTs and EHR data with complex missing data patterns in key outcome or other variables while effectively accounting for complexities within the data structure, minimizing bias, and maintaining reasonable efficiency. In our first project, we developed a framework for personalized treatment strategies when data is subject to missingness, motivated by the Social Incentives to Encourage Physical Activity and Understand Predictors (STEP UP) randomized trial. Our second and third projects examined estimation of average treatment effects using EHR in the presence of misclassified outcomes, selection bias, or missingness. Conducting target trial emulation lies at the intersection of RCTs and EHR, and we considered the issue of missingness in eligibility criteria under this framework in our third project. These approaches address a range of data quality issues when estimating treatment effects and help demonstrate the benefits of utilizing complex data sources to facilitate clinically meaningful guidance.