Electronic health record (EHR) databases are increasingly used to investigate the effect of medications. When the aim is to answer causal questions surrounding the benefits and harms of medications, a key methodological issue is confounder adjustment. Furthermore, successful mitigation of confounding effects often relies on capturing hard to measure markers of frailty, disease severity or health seeking behaviour. This can be especially hard in this context since these data are not collected for research purposes. The high-dimensional propensity score (HDPS) algorithm is a semi-automated data driven approach for confounder identification, prioritisation and adjustment tailored for use in large healthcare databases. The HDPS is increasingly applied in pharmacoepidemiological studies amid growing evidence supporting the benefit of these approaches in comparison to standard covariate adjustment methods. Developed in administrative claims databases, there has been little exploration of how best to translate the algorithm beyond this setting. In this thesis, I propose modifications for implementing HDPS principles in UK primary care EHRs that aim to better characterise features of these data. These modifications are applied to case studies where residual confounding is a key concern. In addition, I propose diagnostic tools and guidance for the reporting of HDPS approaches in general. Furthermore, the developed HDPS approaches are implemented in the Stata statistical software package. Finally, I extend the existing HDPS framework to support the incorporation of laboratory test result information. Whilst the HDPS is not a panacea, the collective findings of this thesis demonstrate the utility of HDPS approaches for overcoming intractable confounding in UK EHRs. Future work will further explore the use of test result data within the HDPS framework.