Curating Electronic Health Record Data to Assess Causal Inference Effect of Metformin on Hypertension Population Progression to Chronic Kidney Disease

Causal inference is a methodology to assess the impact of one variable on another by establishing a cause-and-effect relationship. This paper deals with causal inference analyses using Electronic Health Record (EHR) data from the UK Biobank. Key challenges in this area include data curation, missing data, inconsistencies, and invalid entries, which can affect the reliability of results. In this paper, we focus on dealing with these challenges and illustrate our approach with a case study that examines the effect of a drug, i.e., metformin on the progression to chronic kidney disease (CKD) in patients with hypertension as comorbidity. Our approach involves identifying confounders, including other relevant medications, demographic factors, and clinical characteristics. We use two popular techniques, inverse probability weighting and propensity score matching to evaluate metformin's impact on CKD progression. Our findings highlight the significance of rigorous data preparation and the need for careful methodological choices in conducting causal inference studies. With effective use of EHR data, this paper provides a practical guide for similar analysis, offering an alternative method to understand drug effects and disease progression in clinical research, emphasizing the need to address challenges to avoid misleading conclusions in clinical research.