Uncontrolled confounding is a challenge that may affect the validity and limit the interpretation of the results from observational studies. Such confounders arising from the use of large databases (measured and unmeasured confounders) can be handled through the choice of appropriate study design and analysis method of the data (e.g. stratification, standardization, regression models, propensity scores and combining with machine learning tools with high-dimensional propensity score (hdPS) method).
Additionally, biases can possibly arise from the inappropriate handling of data used to generate RWE such as selection bias and information bias. These biases need to be addressed in order to have valid and reliable results from RWE (e.g. matching patients, target trial emulation).